The women won me over one athlete at a time, and a boy did the rest

I never had much use for sports. Growing up, what I saw was men being macho, and macho made me uncomfortable. Athletes, in my experience, translated into bullies, and the bullies were the ones making my everyday life miserable. So I did the sensible thing a kid does when something hurts. I built distance. A healthy wall between me and sports, and between me and the jocks on the other side of it. By the time I was grown, the wall was just part of the architecture. Sports were what other people cared about.

What slowly got through the wall was the women.

It started young and by accident. I grew up watching tennis and figure skating, usually because the Olympics were on, or because it was Wimbledon fortnight and the television was tuned to it. My parents were golf fans, so every Sunday, a tournament was played in the background. I’ll say this for the PGA: the hushed voices of the announcers gave me some of the best naps of my life. But when the LPGA was on, something was different. I stayed awake. I couldn’t have told you why at the time.

The WNBA started in 1997, and my interest grew like a slow tide, an inch at a time over the years. In 2001, I watched Jennie Finch, probably the most famous softball player who ever lived, put together a perfect 32–0 season at Arizona, the only one in NCAA softball history, throwing no-hitters on the way to a national title, and then go to Athens in 2004 and bring home Olympic gold.

The basketball stayed a slow roll for me right up until Caitlin Clark and Angel Reese met in the 2024 NCAA final, a game that felt like a brawl waiting to happen, all that talent and animosity crammed into forty minutes, and then both of them went straight to the pros.

In 2024, the PWHL finally started playing, after years of false starts for women’s professional hockey. Soccer barely registered for me, with one exception that kept repeating: the United States women’s national team kept winning Olympic gold, in 1996, then 2004, 2008, 2012, and again in 2024. Then I watched “Ted Lasso,” the show about an English men’s football club, and I’ll admit it was great. Did it make me a fan of men’s soccer? It did not. I started tuning in to the NWSL. What did you expect?

I’ll put in a word here for women’s beach volleyball, too. Kerri Walsh Jennings and Misty May-Treanor were two of the best players on the planet for the better part of a decade, winning three straight Olympic golds from 2004 to 2012. I never missed those matches.

Through all of this, I could not have been paid to watch men play sports. With one standing exception: the Super Bowl. For the commercials, obviously.

But there was a second thread running beneath the entire timeline, and it became the exception that actually mattered.

Learning the men’s side

Women’s gymnastics was always there. I watched it now and then, the way you watch something every four years and then forget about it. Then I became a mother, and everything about how I watched it changed. My youngest son wanted to watch gymnastics, so that’s what we did, the two of us, in front of the television. And then that same kid got a little older and decided he didn’t want to be a gymnast. He needed to be one. I mean that exactly. It wasn’t a preference. It was an imperative that he could not be talked out of.

There was a problem. Eastern South Dakota did not have men’s gymnastics coaches. So I changed jobs. I changed my schedule. Then, with my son as my only athlete, I became a gymnastics coach four hours a day, six days a week, and I learned men’s artistic gymnastics from the ground up, the way a coach has to learn it and not the way a fan picks it up from the couch. Six events instead of the women’s four. Floor, pommel horse, still rings, vault, parallel bars, and high bar. I learned why the iron cross on still rings is a feat of raw strength that looks like stillness, why the pommel horse is the event that quietly breaks people, why a clean release-and-catch on high bar should terrify everyone watching and usually doesn’t, because the gymnast made it look easy.

So yes, I’ll admit to an affinity for men’s artistic gymnastics, at any level. If there’s one men’s sport that gets watched in my house, it’s that one. When Stephen Nedoroscik took off his glasses at the 2024 Paris Olympics and hit a pommel horse routine that helped the American men win their first team medal in sixteen years, then took an individual bronze of his own, I wasn’t watching a meme. I was watching the one men’s event I’d spent years learning to read, finally getting seen. And I’m still coaching today, ten years later, working with little boys learning the art of controlling their bodies and defying gravity, though not as many hours a week as I once put in.

That’s the exception. It’s specific, it’s personal, and it’s earned. Everything else points the other way, and it’s worth being honest about why.

What I walked away from

Here is the part that connects back to that wall I built as a kid. When I look at men’s professional sports now, I see the grown-up version of exactly what I walked away from. The macho posturing. The entitlement. The sense that the games exist to give a certain kind of man a stage. I’m not going to dress that up as a clean statistic. The honest version is narrower than the slogan. Studies of arrest records actually find that NFL players, for one, get arrested at lower rates than the national average for men their age. But certain categories, domestic violence among them, were a problem the leagues spent years refusing to take seriously, and the culture that allowed that is the same culture I recognized on the playground a long time ago. I don’t enjoy watching it. After everything, I think that’s allowed to be a reason.

There is a second reason, and it’s even more honest. I will never reach the athletic level of an elite man, and I know exactly why. The performance gap between top men and top women is real and well measured. It runs around ten to twelve percent across most running events, a little narrower in the shortest sprints and wider at the longest distances. Men carry more muscle, denser bone, higher testosterone, and a bigger engine: elite women’s peak oxygen uptake runs roughly ten to twenty percent below men’s, depending on the sport. None of that is in dispute, and I have no interest in pretending it away.

What they do with it

Here’s the turn. That gap is the whole reason the women move me. When the ceiling was lower, and the resources were thinner for most of the sport’s history, every inch of mastery had to be fought for against the grain. Men’s excellence, I can admire the way I admire a tall building. Women’s excellence. I admire the way I admire someone who built the building after being told for a century she couldn’t hold a hammer.

Look at what they do with it.

Lindsey Vonn won four overall World Cup titles and an Olympic downhill gold, the first ever for an American woman, and retired in 2019 with 82 World Cup victories, the most by any woman at the time. In 2012, she asked to race a men’s World Cup downhill. The international federation said no, ruling that one gender isn’t entitled to enter the other’s races. She was candid that she’d be only averagely competitive against the men. The point was never to win. The point was to measure herself against the fastest skiers alive.

When she finally skied the men’s Streif course at Kitzbühel in 2023, the run that terrifies the men who race it for time, she came down and said she had even more respect for them now. Then she returned to ski racing at 40, after a partial knee replacement. At forty, she became the oldest woman ever to reach a World Cup podium, and the following season, at forty-one, she won a World Cup downhill outright, the oldest racer of any gender ever to win one. That is what working for every advantage looks like.

Simone Biles is the most decorated gymnast in history, with 41 Olympic and World Championship medals between them, and five skills named after her because no one else will attempt them. The federation that governs her sport has at times deliberately scored her hardest skills lower than their difficulty warrants, openly, to keep other women from breaking themselves trying to copy her.

Think about what that means. She is so far ahead that the rulebook treats her as a hazard. In Tokyo, she pulled herself out of finals when she lost her air awareness in the middle of a twisting vault, a genuinely dangerous thing, and took the public heat for putting her body and mind first. Then she took two years away, came back, and won three more golds in Paris. The men’s side of her sport is the one I coach. I watch her side because I’m seeing the outer limit of what a human body can be taught to do.

The whole history in one paycheck

And then there’s Caitlin Clark again, because her story is the one that shows you the machinery at work. She left Iowa as the all-time leading scorer in NCAA Division I basketball, women’s or men’s, with 3,951 points. The 2024 national championship game she played in drew 18.9 million viewers and outdrew the men’s final for the first time in history.

When the Indiana Fever made her the number one pick, her rookie salary was $76,535. The man taken first in that year’s NBA draft signed for a first-year salary north of $12.5 million. One finance professor’s analysis credited Clark with roughly a quarter of the WNBA’s economic activity in her rookie season. She moved the entire market, and she was paid like an intern.

That contrast is the whole history of women’s sports compressed into one paycheck. The talent was always there. The investment wasn’t made, and the missing investment was used as proof that demand didn’t exist, which justified withholding the investment all over again. It was a closed loop, and the women were trapped inside it.

The loop is finally breaking, and you can watch it in real numbers. In 2025, Clark missed much of the season due to injury, and the WNBA’s audience held steady while its attendance climbed to a record 3.15 million. The boom turned out to be bigger than any one player. In early 2026, the league signed a new labor deal that ties salaries to revenue for the first time in the history of women’s professional sports, raised the team salary cap from about $1.5 million to $7 million, and pushed the average salary from roughly $120,000 toward $583,000. Clark’s own salary jumped from that $76,535 to around half a million. The pioneers are starting to get paid as though the thing they built has value, because it does.

I won’t oversell the case that the women’s game is simply better to watch, because part of that is taste. Fans talk about teamwork over isolation, fundamentals over showboating, and a style closer to what the games were invented for. Some of that is real, and some of it is just what I happen to like. But the parts that aren’t subjective are striking.

Women’s sports fans are the most engaged audience in sports by some measures. Tickets are affordable. The fan culture is warm in a way that the men’s leagues, with their corporate boxes and their swagger, mostly stopped being a long time ago. The 2023 Women’s World Cup drew nearly two million fans in person. Women’s elite sport crossed a billion dollars in annual revenue for the first time in 2024. This is not charity. People are showing up.

How recent all of this is

But the thing that organizes all of it, the fact I keep circling back to, is how recent this is. Title IX became law on June 23, 1972. It’s younger than plenty of people who are still going to work every morning. The law itself says nothing about sports. It only says that no one can be shut out of an educational program because of their sex. In 1971, the year before it passed, about 295,000 girls played high school sports in this country, roughly one in twenty-seven. Today, it’s about 3.4 million, roughly two in five. The women I idolize are the second generation, maybe the third, to grow up legally entitled to a place on the field.

That history isn’t finished, and I won’t pretend it is. Title IX never required equal money, and the money still isn’t equal. Fifty years after the law was passed, girls were being offered fewer athletic opportunities than boys had been before the law existed. The gap moved. It didn’t close.

But that’s the point. The excellence I’m watching is brand new. Jennie Finch’s perfect season, Lindsey Vonn outracing her own age, Simone Biles rewriting what a body can do, Caitlin Clark out-drawing the men who were supposed to be the main event, all of it is happening inside a window that opened in my own lifetime, against a head start the men got measured in centuries. I don’t watch the women because the men aren’t good. I watch them because they have to earn, in fifty years, the standing the men inherited, and they’re doing it right in front of us, and I don’t want to look away.

The wall I built as a kid is still there, mostly. I still can’t make myself care about the men’s leagues that grew out of the playground I couldn’t wait to leave. But two things got through it. The women, who won me over one athlete at a time, until I was the person staying awake for the LPGA and never missing a beach volleyball match. And a small boy who decided he needed to fly, in the one men’s sport I’ll always make room for, because I’m the one who taught it to him.

I watch everything else on the women’s side because they taught it to themselves.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

For more stories about the barriers women break and the journeys that make women’s victories even more meaningful, follow Fourth Wave. Also check out our Livestreams and sign up for our Newsletter. Let’s grow the community! Have you got a story or poem that focuses on women or other targeted groups? Submit to the Wave!

On Sunday mornings, I run the soundboard at a little country church outside Sioux Falls. I’m not always sure I believe what comes through the speakers. I show up anyway.

So when someone announces that a stranger’s Christianity is fake, my ears prick up. I’ve spent many years wondering about my own.

And if you have faith but you’ve quietly given up on church, or you still sit in a pew arguing in your head the way I do, you are in an enormous crowd. Gallup found that for the first time in eighty years of asking, fewer than half of Americans belong to a congregation at all. Most of us are out here somewhere, holding something we can’t quite set down and can’t quite carry through the front doors.

Which brings me to James Talarico, Texas’ Democratic candidate for the U.S. Senate.

The sentence, and the half they kept

Talarico is a Texas state representative running for the United States Senate, and he is studying to be a Presbyterian minister. A few days ago, a clip from an old podcast went around, and a good part of the internet decided he hates God. In it, he says, I always think of myself as a Christian who hates Christianity.

That is a real quote. Here is the part that the clip cuts off.

He kept talking. He said he keeps getting pulled back to the faith. Nowhere else, he said, in no philosophy or economic theory, does he find anything as radical as the teachings of that barefoot rabbi. The hatred is for the machinery, for the kind of church that climbs into bed with politics. The love is for the man at the center of it.

If you have ever loved something and been furious at what people turned it into, you already understand the whole sentence. They kept the angry half and threw away the half that explained it.

You can argue with Talarico’s theology. Plenty of thoughtful Christians do. He has said things about God and gender and other religions that land hard on traditional ears, and that’s a fair fight, and an old one. But being wrong about doctrine is not the same as hating God. Half the people in any sanctuary would flunk that test, and most of the saints with them.

So set the theology aside. I want to talk about the people holding the measuring stick.

Now look at the accusers.

The clip didn’t climb out of an archive on its own. It was handed to the press by a party operation in the middle of a dead-even Senate race, the same week a poll showed the two men tied. The timing is the tell. This was never theology. It was opposition research with a cross stuck on top.

So look at who lined up to throw it.

His opponent, Attorney General Ken Paxton, went on television and pronounced that Talarico doesn’t understand Christianity in any form or fashion. This is the same Ken Paxton who was indicted for securities fraud and impeached by his own party’s House before the Texas Senate voted to let him off. A man under that particular cloud, grading a divinity student’s walk with God.

Senator Ted Cruz declared that Texas will never elect someone who thinks God is nonbinary. The Republican National Committee reached into the Sermon on the Mount, pulled out wolf in sheep’s clothing, and hurled the verse at a man to win an election.

Picture that for a second. People who handle Scripture like a nightstick, lecturing a seminary student about whether he respects it.

Who is really emptying the pews?

Here is the part none of them will say from a stage.

The one time Jesus actually stopped to describe the final judgment, he did not hand anyone a theology quiz. He asked whether you fed the hungry, whether you took in the stranger, and whether you visited those in prison. That was the entire exam. And the officials now auditing Talarico’s faith are, more often than not, the same ones who cut the programs that feed the hungry and call the stranger an invader.

So, who looks more like the Christian here? The seminarian who keeps showing up for a faith that embarrasses him in front of his own friends, or the officeholder who wears it like a campaign button and governs as if the Sermon on the Mount were a clerical error.

I am not going to pretend that it is a close call.

And the performance is costing the church the very people it claims to want. When Pew asked Americans why they walked away from the religion they grew up in, roughly a third named scandals among religious leaders, and roughly a third named the faith’s politics. Read that again. The loudest defenders of Christianity are its worst advertisement. They are not guarding the faith against James Talarico. They are the reason the little building I sit in on Sundays keeps getting quieter.

If you left all of this behind, tired of people who shout about Jesus and act nothing like him, I want you to hear something. Your instinct was not a betrayal. It was eyesight. My own argument has never been with God. It has been with what people do with God, which is mostly to file him down into a weapon and go looking for someone to swing him at.

They questioned a man’s faith on a few points in a poll. If you ever want to know whose faith is the real thing, skip the press releases and watch their hands. Watch who actually feeds somebody.

That was always the only test.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

References

Gallup. “U.S. Church Membership Falls Below Majority for First Time.” 2021. https://news.gallup.com/poll/341963/church-membership-falls-below-majority-first-time.aspx

Guo, Kayla. “Ken Paxton and James Talarico Are Neck and Neck in U.S. Senate Race, New Poll Finds.” The Texas Tribune, June 23, 2026. https://www.texastribune.org/2026/06/23/texas-us-senate-poll-ken-paxton-james-talarico-ut-austin/

Pew Research Center. “Why Do Some Americans Leave Their Religion While Others Stay?” December 15, 2025. https://www.pewresearch.org/religion/2025/12/15/why-do-some-americans-leave-their-religion-while-others-stay/

Pierce, Tony. “Austin’s James Talarico Is Running for U.S. Senate.” KUT (Austin’s NPR Station), September 9, 2025. https://www.kut.org/politics/2025-09-09/james-talarico-austin-texas-u-s-senate-2026-election-democrat

“Talarico Leads Paxton in Texas Poll.” The Hill, June 10, 2026. https://thehill.com/homenews/campaign/5916384-talarico-leads-paxton-texas-poll/

“James Talarico Leads Ken Paxton in Texas Senate Race Poll After GOP Runoff.” The Hill, May 30, 2026. https://thehill.com/homenews/campaign/5902035-texas-senate-race-poll/

D’Abrosca, Peter. “Texas Senate Candidate James Talarico Said He ‘Hates Christianity’ in 2021.” Fox News, June 24, 2026. https://www.foxnews.com/politics/talarico-says-hates-christianity-unearthed-interview-transgender-latinx-theologian

“Texas Senate Hopeful Talarico Under Fire Over Resurfaced ‘Christian Who Hates Christianity’ Remarks.” The Washington Times, June 23, 2026. https://www.washingtontimes.com/news/2026/jun/23/texas-senate-hopeful-talarico-fire-resurfaced-christian-who-hates/

I was confirmed five weeks before my fourteenth birthday. I stood at the front of a Lutheran sanctuary in South Dakota and affirmed, in front of the whole congregation, a faith I was told I now understood well enough to claim as my own. Substitutionary atonement. The resurrection. Eternal life, and its opposite. I had taken two years of confirmation class to get there, and the adults in that room agreed I was ready.

I was thirteen.

Here is the part nobody in that sanctuary thought was strange. The same culture that trusted thirteen-year-old me to comprehend the machinery of eternal salvation, to weigh sin and grace and hell and choose, would have told that same thirteen-year-old, had I said something about my own gender or about a classmate’s, that I was far too young to know anything of the kind. Too young for that. Old enough for this.

The writer Emily Joy Allison put the whole contradiction in a single viral post: it’s “so weird how children are too young to know anything about their own gender or orientation, but old enough to understand eternal conscious torment and substitutionary atonement and make a lifelong permanent commitment to Christianity. Someone should look into that.”

So let’s look into it. Not to score a point on the church I still belong to, but because the contradiction is real, it’s doing real harm to real kids, and the words being thrown around to defend it, grooming and indoctrination, mean something specific. They just don’t mean what the people shouting them think they mean.

The order is backward.

Start with what developmental psychology actually says, because the science doesn’t split the way the slogans need it to.

Children know their own gender early. The cognitive-developmental sequence Lawrence Kohlberg laid out in 1966, and which has held up across cultures since, runs like this: a child labels their own gender at two or three, understands it’s stable over time at four or five, and reaches full gender constancy somewhere between five and seven. Kristina Olson’s TransYouth Project, the largest longitudinal study of socially transitioned trans children, found that their sense of their own gender looks like that of cisgender kids: same strength, same pattern. I didn’t need a study to tell me this. I knew I was a girl in kindergarten. The boys in my class knew they were boys, and nobody asked them to prove it.

Awareness of sexual orientation arrives a little later, around the edge of puberty. Drawing on a national probability sample, Bishop and colleagues (2020) found the average age of first same-sex attraction was 11.75, with some reporting it as early as eight. Pew’s 2013 survey of LGBT adults found a median age of twelve for first feeling something other than straight. Earlier work tied that first spark to the hormonal changes of around age ten.

Now hold that against the cognitive equipment required actually to understand the doctrines I affirmed at confirmation.

Abstract theological reasoning, the kind that can hold a concept with no physical referent and turn it over, is what Jean Piaget called formal operational thought, and it doesn’t even begin to come online until around age eleven or twelve. Plenty of adults never use it consistently. And these aren’t gentle abstractions. Eternal conscious torment. Penal substitution is the idea that an innocent man was punished in the place of the guilty to satisfy divine justice. A child’s understanding of plain death, far more concrete than any of that, isn’t typically complete until ages eight to ten. One study of children’s afterlife beliefs found that by age ten, their explanations mixed contradictory biological and supernatural ideas, and that those beliefs tracked their parents’ beliefs. Which is the tell, and I will come back to it.

Line the ages up, and the double standard inverts itself:

1. A child knows their own gender by five to seven. (some say as early as three)

2. A child becomes aware of their orientation around ten to twelve.

3. A child can’t genuinely comprehend substitutionary atonement or eternal torment until eleven or twelve at the earliest, if ever.

The thing we trust children to understand is the hardest one. The things we tell them they are too young to know are the ones they figure out first. The developmental order is the opposite of the political talking point.

The words mean something.

This is where I want to slow down, because two words are carrying the entire weight of the backlash against LGBTQ kids, and both have real definitions that the shouting has buried.

Grooming isn’t a metaphor. It’s a forensic term for a specific process. The forensic psychiatrist Michael Welner describes it as the way an offender draws a particular child into a sexual relationship and keeps that relationship secret. He is clear that secrecy is the whole point. Child-protection groups break it into stages: an adult selects a specific child, builds trust with the child and often with the parents and the wider community, isolates the child, sexualizes the relationship, and uses secrecy and control to keep it hidden. A targeted child. Concealment. The goal of abuse.

Read that description again and ask what it actually fits. An authority figure, trusted by the community, uses secrecy and emotional pressure to override a child’s developing sense of self. That isn’t a teenager saying I think I’m gay to a school counselor in the open. It’s a much older and more documented pattern, and a great deal of it happened inside churches. Allison knows that world; she helped launch #ChurchToo in 2017 to surface exactly that kind of abuse in Christian institutions.

The word got unhooked from its meaning on purpose, and you can date it. When Florida passed its Parental Rights in Education law in 2022, the governor’s press secretary tweeted that anyone opposing the bill was “probably a groomer.” A report from the Human Rights Campaign and the Center for Countering Digital Hate found that in the month after the bill passed, the average number of tweets per day using slurs like “groomer” against LGBTQ people jumped 406 percent. A year later, an ADL and GLAAD report counted hundreds of anti-LGBTQ incidents, with the false groomer-and-pedophile smear the single most common theme. A precise word for real horror was repurposed as a slur aimed at people for existing .n public.

Indoctrination has a definition, too. In the philosophy of education, it means instilling beliefs in a way that leaves the person unwilling or unable to question them later, beliefs held without regard to evidence and sealed against challenge. Real education, the philosopher Harvey Siegel argues, may plant beliefs in a child before the child can fully justify them, but it does so to equip the child with the tools to question those beliefs down the road. The difference is the door. Indoctrination locks it. Education leaves it open.

By that standard, telling a child she faces unending torment unless she accepts a particular doctrine, and discouraging her from questioning it, sits much closer to the textbook definition of indoctrination than an age-appropriate lesson that gay and trans people exist. Which is the one routinely branded “indoctrination” by people who pair the word, without apparent irony, with “groomer.”

Mostly, it gets handed down.

There is one more piece of evidence that deflates the idea of the freely choosing child believer, and it isn’t subtle.

Religion is inherited far more than it’s chosen. Pew’s analysis of a 2019 survey of parent-teen pairs found that 82 percent of Protestant parents had teens who also identified as Protestant, 81 percent of Catholic parents had Catholic teens, and 86 percent of religiously unaffiliated parents had unaffiliated teens. The single best predictor of a thirteen-year-old’s religion is the religion of the adults who raised them. And the commitment lands early: the evangelical Barna Group has long reported that a large majority of Americans who become Christians do so before they turn eighteen, most of them before thirteen. Their founder summarized that what you believe by thirteen is largely what you will die believing.

I’m not mocking that. I lived it. I was baptized at seven weeks old, before I could hold up my own head, let alone consent to anything. That is simply how my tradition does it, and I will defend its logic in a moment. But let’s not pretend a thirteen-year-old at the front of a sanctuary is a free agent weighing Reformation theology on the merits and independently electing into it. She is, overwhelmingly, doing what the adults she loves and depends on have raised her to do. That’s not a scandal. It’s just true, and any honest account has to say so.

The strongest argument against me, said fairly.

If I am going to argue this, I owe the other side its best case, not a strawman. There are two real objections, and one of them is genuinely good.

The good one is about baptism. In my own Lutheran tradition, and in the Catholic one, infant baptism is precisely not a request for the infant to understand or commit anything. The whole point, as the Augsburg Confession put it back in 1530 and as the ELCA still teaches, is that God is the one doing the work, not the baby. The parents and godparents make the promises. Confessional Lutherans actively reject the idea that salvation hangs on a child’s rational decision. So when Allison says children are made to “make a lifelong permanent commitment,” a Lutheran can fairly answer: not at the font, they aren’t. Nobody asked seven-week-old me to comprehend a thing.

That’s correct. And it doesn’t rescue the contradiction. It relocates it and lands it harder.

Because my tradition doesn’t leave it at infant baptism, it holds the conscious, personal yes in reserve. It asks for it later, at confirmation, which the ELCA formally calls the Affirmation of Baptism. And confirmation in the Lutheran world typically occurs around age fourteen, after a couple of years of catechism. I was thirteen. So the tradition itself locates the moment of genuine religious commitment at roughly fourteen, which is older than the age at which kids become aware of their orientation and well within the age range that the same communities insist is far too young to know anything about gender or sexuality. The baptism objection, followed honestly to its end, doesn’t dissolve the double standard. It dates and confirms it.

The second objection is a category argument: accepting a doctrine taught by an authority and discovering something true about yourself are different kinds of mental acts, so the comparison is unfair. That’s right too, and it also cuts the wrong way for the people who make it. Knowing your own gender or noticing who you are drawn to is something you find, often without wanting to. Allison has described realizing she was attracted to women as a light coming on that she couldn’t switch back off. Discovering a fact about yourself takes less abstract machinery than evaluating a metaphysical claim about a substitute sacrifice, not more. If the two really are different categories, then children are better positioned to know their own hearts than to adjudicate the doctrine of hell, which is the opposite of what the objection needs to be true.

What I actually think is going on

I should be honest about where I stand, since I have been circling it.

My skepticism isn’t aimed at God or the idea of God. It’s aimed at what human beings have done with God. I still belong to my church. I still run the sound booth at a little country congregation outside Sioux Falls, and I trained the teenage girls who run the gear while I roam the building, checking the speakers. I love those people. But I have read too much history to believe that most of what gets handed down from a pulpit fell from the sky. Men built a great deal of it, and much of it was built to keep men in charge.

The faith that once refused to bow to Caesar became, under Constantine and his successors, the official religion of the empire, its creeds hammered out under imperial sponsorship, and its rivals criminalized within a generation. Creed, sword, and empire. The institution learned early that controlling what people believe and starting young are how you hold power over them. Teaching a frightened child that hell waits for the wrong answer isn’t an accident of doctrine. It’s a feature, and it works.

So here’s what I can’t make square. The instinct to protect children is real and good. I share it. But watch where it points. It doesn’t flinch at handing a seven-year-old the concept of eternal conscious torment, or asking a thirteen-year-old to commit her one life to a system she’s developmentally years away from being able to question. It flinches, hard, only when a child says something true and inconvenient about who she is.

That’s not a concern for the child’s mind. If it were, it would worry first about the hardest thing we ask children to swallow, not the easiest thing they figure out on their own. It’s a concern for control. It always was.

I was old enough, at thirteen, to be trusted with hell. A gay kid in that same pew was supposedly too young to be trusted with themself. Someone really should look into that.

I think we just did.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

A man said a sentence, and a machine cut it in half so the surviving half would mean the opposite of the whole. The man is James Talarico, the Democratic nominee for one of Texas’s United States Senate seats, and for three days his sentence has been traveling the right-wing internet in its amputated form.

Here is what got cut. In a March 2021 episode of the Activist Theology Podcast, one of the hosts asked Talarico a long question about whether seminary, “still a dominant system,” was worth entering, and how a person opts out of the systems they get “conscripted into.” Talarico answered: “I always think of myself as a Christian who hates Christianity. And, like, I always get drawn back into it because nowhere else, in no other political philosophy, in no other economic theory, do I find anything as truly radical or revolutionary as the teachings of that barefoot rabbi.” The full audio has been published by Fox News and transcribed in full by The Post Millennial, among others.

Read the whole thing and the meaning is not hidden. He was answering a question about the “dominant system” of organized religion, and he said the thing every honest believer who has ever been let down by a church has said in some form: I cannot stand what this has become, and I cannot leave it; the man at the center of it will not let me go. The Christianity he hates is the system the host had just named. The faith he cannot quit is the barefoot rabbi.

How the sentence traveled

Now here is what the machine kept. A clip of the first eleven words, stripped of the question that prompted them and the clause that finished them, posted by the Republican National Committee’s research account, picked up the same day by The Federalist as an “exclusive,” then by The Daily Wire, Fox News, Breitbart, the Christian Post, and a column of aggregators, all on June 23 and 24, 2026. The version that landed on Matt Walsh’s Facebook page, the post that probably brought you here, opens with a line that no longer belongs to Talarico at all: that he is “admitting his disdain for biblical Christianity” and pushing a “fringe faith worldview across the Lone Star State.” Those are not his words. They are the Daily Wire’s lede, the gloss the machine bolted on top of the cut.

That is the turn. A sentence about loving Jesus and distrusting the church became a confession of contempt, and it became one by deletion.

So: is it true? Does James Talarico hate Christianity?

Who Talarico is

He trained at Austin Presbyterian Theological Seminary, where he earned a master’s in theology, and the June 2026 coverage keeps noting, almost as an indictment, that he is still a Presbyterian seminarian on his way to the pulpit (KUT, citing the Texas Tribune). His minister of more than thirty-five years, Jim Rigby, runs a famously progressive congregation at St. Andrew’s Presbyterian in Austin. When Texas moved in 2023 to hang the Ten Commandments in every public school classroom, Talarico stood up in committee and called it out: “a religion that has to force people to put up a poster to prove its legitimacy is a dead religion, and it’s not one I want to be a part of” (KUT). His two recurring texts are the two commandments Jesus named, love God and love your neighbor, and the judgment scene in Matthew 25 where the only question is whether you fed the hungry and visited the prisoner. This is not an exotic theology. It is mainline Protestantism, the kind preached from ten thousand American pulpits for a century.

Where his theology is debated

I am not here to baptize everything the man believes. Talarico has said things from the floor and the microphone that serious people, including serious Christians, find genuinely hard to swallow: that God is “nonbinary,” that his support for abortion rights follows from reading the Annunciation as Mary giving consent, that the teachings of Jesus look “a lot like the teachings of the Buddha and other mystical traditions.” First Things, no one’s idea of a left-wing outlet, ran a whole essay calling his theology backward. You can think his Christianity is wrong. You can think it is muddled. Those are real arguments, and they are arguments about doctrine, which Christians have been having with each other since the first century. Not one of them is the same thing as hating Christianity. A person can be wrong about the Trinity and still love God. The church has always been full of such people. So have its pulpits.

Believers who hated the church

Here is the part the machine is counting on you not knowing. The believer who hates the church is not a modern progressive invention. He is one of the oldest figures in the whole tradition.

Open the book of Amos, eighth century before Christ, and you find God himself saying it, in the first person, about his own people’s worship: “I hate, I despise your festivals, and I take no delight in your solemn assemblies. … Take away from me the noise of your songs; I will not listen to the melody of your harps. But let justice roll down like waters” (Amos 5:21–24, NRSV). That is contempt for religious performance, spoken inside the covenant, by the God the performance was supposedly for.

Open Kierkegaard, the Lutheran’s Lutheran, who spent the last two years of his life, 1854 and 1855, publishing a serial assault on the entire Danish state church under the title Attack Upon Christendom, on the argument that a comfortable, official, everyone-is-automatically-a-Christian religion had quietly abolished the real thing. He was not an unbeliever sneaking into the choir. He was a Christian who had concluded that the official church had become the enemy of faith.

Open the most quoted American Sermon of the twentieth century. In the Letter from Birmingham Jail, dated April 16, 1963, Martin Luther King wrote, “I have been so greatly disappointed with the white church and its leadership,” and then, two sentences later, told you exactly what kind of disappointment it was: “I say this as a minister of the gospel, who loves the church; who was nurtured in its bosom … and who will remain true to it as long as the cord of life shall lengthen.” There can be no deep disappointment, he wrote, where there is not deep love.

Put Talarico’s sentence next to those three and the structure is identical. Hatred of the established church; devotion to the thing it was built to carry. He even handed you the second clause for free. He told you why he keeps coming back. The machine cut that part off. The second clause is the whole defense.

The accusation is the weapon.

Now watch the move at the center of this. It is the one I keep coming back to. The people who cut the sentence are not defending Christianity. They are using it.

The clip was released by a party research operation in the middle of a tied Senate race. On June 23, 2026, the same Tuesday The Federalist ran its “exclusive,” the University of Texas’s Texas Politics Project released a poll showing Ken Paxton ahead of Talarico by a single point, 43 to 42, inside the margin of error, after Paxton had gained nine points on him since April (Texas Tribune). Paxton, who is himself running for the Senate after his own party’s House impeached him in 2023, had already gone on Fox the week before and pronounced, of a Presbyterian seminarian, “I don’t think he understands Christianity in any form or fashion” (The Hill). The RNC framing called Talarico “a wolf in sheep’s clothing.” Senator Ted Cruz announced that “Texas will never elect someone who thinks God is nonbinary.”

“Wolf in sheep’s clothing” is Matthew 7:15. They reached into the Sermon on the Mount, pulled out a verse, and aimed it at a man for votes. That is Scripture used as a closed weapon, which is the exact thing they are accusing him of doing. “Biblical Christianity,” in the post you read, is not a description of anyone’s actual belief. It is a purity standard, defined by the people administering it, applied to a candidate three weeks after the general-election field was set, calibrated to cost him the moderate Christians the poll says he needs.

That is the turn under the turn. The accusation of weaponizing faith is itself the weapon. And it is a more dishonest weapon than anything in the clip. It does not even misquote him; it deletes him and writes new words in his mouth.

Where I stand

I should tell you where I am standing. I am not a Talarico voter; I do not live in Texas, and I would not call myself a believer in the way he is one. I am a skeptic who still runs the sound booth at a small country church on Sunday mornings outside Sioux Falls, which means I spend a fair part of my week inside a building I have a complicated relationship with, listening to a faith I argue with in my head. My skepticism has never been aimed at God. It has always been aimed at what human beings do with God, which is mostly to whittle him into a stick and hit each other with it.

So I know Talarico’s sentence. I know it the way you know your own handwriting. A Christian who hates Christianity and cannot leave it is not a scandal. It is half the people I have ever sat next to in a pew, and it is the more honest half. The dishonest half is the one who holds a verse like a knife and calls it faith.

They cut his sentence in half. Read the other half. He told you he could not quit the barefoot rabbi. He was telling the truth.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

References

Cassidy, Julia. “Talarico Offends Christians Yet Again by Claiming He ‘Hates’ Christianity.” Townhall, June 23, 2026. https://townhall.com/tipsheet/julia-cassidy/2026/06/23/talarico-calls-himself-a-christian-who-hates-christianity-n2678164

D’Abrosca, Peter. “Texas Senate Candidate James Talarico Said He ‘Hates Christianity’ in 2021.” Fox News, June 24, 2026. https://www.foxnews.com/politics/talarico-says-hates-christianity-unearthed-interview-transgender-latinx-theologian

Guo, Kayla. “Ken Paxton and James Talarico Are Neck and Neck in U.S. Senate Race, New Poll Finds.” The Texas Tribune, June 23, 2026. https://www.texastribune.org/2026/06/23/texas-us-senate-poll-ken-paxton-james-talarico-ut-austin/

King, Martin Luther, Jr. “Letter from Birmingham Jail.” April 16, 1963.

https://letterfromjail.com/

“Navigating Love & Legislature: A Conversation with Rep. James Talarico.” Activist Theology Podcast, March 2021 (transcript excerpts, The Post Millennial). https://thepostmillennial.com/james-talarico-says-he-hates-christianity-in-interview-with-transqueer-theologian

New Revised Standard Version Bible. Amos 5:21–24. National Council of the Churches of Christ, 1989. https://www.biblestudytools.com/nrs/amos/passage/?q=amos+5:21-24

Pierce, Tony. “Austin’s James Talarico Is Running for U.S. Senate.” KUT (Austin’s NPR Station), September 9, 2025. https://www.kut.org/politics/2025-09-09/james-talarico-austin-texas-u-s-senate-2026-election-democrat

Reynolds, Joel. “James Talarico’s Backward Christianity.” First Things, 2026. https://firstthings.com/james-talaricos-backward-christianity/

“Texas Senate Hopeful Talarico Under Fire Over Resurfaced ‘Christian Who Hates Christianity’ Remarks.” The Washington Times, June 23, 2026. https://www.washingtontimes.com/news/2026/jun/23/texas-senate-hopeful-talarico-fire-resurfaced-christian-who-hates/

“Talarico Leads Paxton in Texas Poll.” The Hill, June 10, 2026. https://thehill.com/homenews/campaign/5916384-talarico-leads-paxton-texas-poll/

Wheeler, Jacob. “Talarico Says He Hates Christianity in Unearthed Audio.” The Daily Wire, June 23, 2026. https://www.dailywire.com/news/the-audio-thats-blowing-a-hole-in-talaricos-christian-persona

Kierkegaard, Søren. Attack Upon Christendom. 1854–1855 (the original Danish polemic against the established state church).

Ask a chatbot which Supreme Court justice wrote the dissent in Bush v. Gore. There is a strong chance you will get a fluent, confident, citation-shaped answer. There is some chance the answer will be wrong. There is no chance the chatbot knows which case you are in, since the chatbot never asks itself that question.

Last week we opened the engine. The model’s entire job is to pick the next token from a probability distribution, then the next, then the next, until it stops. This week we look closely at what that probability is. The short version of this lesson is: the number is a guess about pattern-match against training data. It is not a guess about truth. The two are usually correlated. They are not the same. Where they pull apart, the system produces fluent confident error. That gap, owned at the mechanism level here, is the conceptual heart of hallucination, which we will name and unpack starting in Lesson 15.

What the model’s confidence actually measures

When the model assigns a high probability to a candidate next token, it is making a statement about its training data. Roughly: in text patterns like this one, that token tended to be the right next one. That is what the probability is. It is a measure of how confidently the system can fit the present input into the patterns it has previously seen. It is not a measure of whether the resulting sentence will be true.

For most everyday questions, those two things line up. Banana bread recipes tend to be true in the same way they tend to follow predictable patterns; the model’s high-probability completion use ripe bananas is both plausible and correct. The correlation is strong enough that the system feels useful most of the time. The trouble is what happens at the gap.

Ask the system a question where the truth requires looking up a specific fact, and the prediction loop does what it always does: it picks the most plausible next token. Bush v. Gore sounds like a case. The dissent slot, in case-citation patterns, gets filled by a justice’s name. The model has seen many case citations. It will produce a fluent name. Whether the specific name it produces is the right one depends on whether that exact pattern is strong in its training data and whether nothing else has displaced it. Sometimes the answer comes back right. Sometimes the answer comes back wrong, with the same fluency and the same surface confidence as the correct version.

The gap is structural. The loop was never built to check the answer. It was built to predict the next plausible token. Plausible and true are different categories, and the model is operating on the first.

Sampling: even confidence is variable

There is one more wrinkle worth holding in mind. The model does not always pick the highest-probability candidate. Most consumer chat products use sampling: the model rolls dice, weighted toward the high-probability tokens but with room for surprise. Two effects follow.

The first is that two identical prompts in two fresh sessions can produce two different answers. Sometimes both correct. Sometimes one right and one wrong. Sometimes both wrong in different ways. The variability is not noise in the engine. It is the engine working as designed; the design includes a roll of the dice at every step.

The second is that the system can produce a high-confidence-sounding answer that the model itself, on a different roll, would have rejected. There is no internal consistency check across sessions, since each session is a fresh loop starting from the same probability mechanism with a new sequence of rolls.

This means that a confident answer this morning and a contradicting confident answer this afternoon are both genuine outputs of the same machine. The morning answer is not more authentic than the afternoon answer. They are both the system doing its job. The system’s job does not include picking the one that is true.

Three consequences worth holding

A few specific patterns fall out of probability-not-truth that you will meet repeatedly for the rest of this course.

Fluent surface, hollow inside. When the system invents a fact, the invented fact is delivered in the same prose register as the real facts. The court case that does not exist will be cited with a plausible docket number, a plausible judge, and a plausible paragraph of legal analysis. The recipe that does not work will be written in the same warm tone as the recipe that does. The book that was never written will be attributed to the author who never wrote it with the same fluency the system uses for the author’s real books. Fluency is not a signal of correctness. Fluency is the constant background; correctness is the variable on top. This is the inversion of the cue most readers carry from talking with humans, where halting fluency often signals uncertainty. The system has no halting mode. It is fluent always.

Plausible specific is more dangerous than implausible specific. The system rarely invents wildly unlikely facts, since the training data does not support them. The system invents plausibly specific facts, since those are what the prediction loop produces from the probability distribution it learned. Justice O’Connor wrote the dissent is more likely to be the kind of error you get than Justice Snowflakeisplover wrote the dissent, since the first is statistically near the case-citation neighborhood and the second is not. This is exactly the kind of error that survives a quick read.

The model does not know that it does not know. The prediction loop produces a probability distribution every time, including times when it should not have an answer at all. If you ask the model what was on the cover of a magazine that was never published, the loop does not stop and say “I have no information about a magazine that does not exist.” It produces a fluent description of a plausible cover. The system has no internal I-don’t-know signal that consistently fires at the right times. There is research on getting models to flag when they are out of their depth, which we will touch on in Going Deeper. The honest summary is: the research helps, sometimes; it does not fix the underlying problem; the underlying problem is the loop itself.

Why this is the conceptual heart of hallucination

Most of the next several weeks build out of this lesson.

Lesson 15, Hallucination: Why It Lies With a Straight Face, names the phenomenon directly and gives you the practical verification habits that turn this lesson’s diagnosis into a usable defense against the surface-look of fluent confident error.

Lesson 26, How to Catch a Hallucination, hands you a three-move workflow for detecting fabricated content before it does damage.

Lesson 43, Hallucination: What It Is and What It Isn’t, returns to the topic with a working definition and names the two types (intrinsic and extrinsic) that the research literature distinguishes.

All three of those build on what this lesson hands you: the system does not check its work, since its work is to produce plausible tokens, and plausibility was never truth. Once you see the gap clearly, the rest of the hallucination conversation makes structural sense rather than feeling like a list of random failures.

Going Deeper (optional)

There is a small but important research literature on calibration and on whether models can be made to know when they don’t know. Three pieces are worth knowing exist.

Guo, Pleiss, Sun, and Weinberger (2017) showed that modern deep neural networks are systematically overconfident: when a network outputs 90 percent probability for a class, the actual frequency of that class being correct, across many such outputs, is often lower than 90 percent. The networks act more sure than they are. This was the canonical empirical result that calibration is a real problem and a default failure mode of large neural networks, not a quirk of any specific architecture. Calibration is the technical term for closing this gap; many techniques have been developed to improve it, none of them solving the fluent-confident-error problem in language generation since calibration is about the probability numbers, not about the truth of the sentences those numbers underlie.

Kadavath and colleagues at Anthropic (2022) ran the most-cited test of whether language models can assess their own confidence. The headline finding, in the paper’s careful wording, is that language models mostly know what they know. The qualifier carries the weight. The models are partially able to flag their own uncertainty when prompted to do so. The capability is imperfect; the failure pattern is not symmetric; and the practical implication is that asking the model are you sure? is a weak defense against confident error, since the system that produced the confident error is the same system you are asking to second-guess it.

The two findings together explain why “just have the model tell you when it doesn’t know” is not the fix that intuition suggests it should be. The system’s confidence number is uncalibrated. Its self-assessment is imperfect. Both are improving. Neither has closed the gap. The gap is structural, and the structural gap is what this lesson opens up at the conceptual level. Lesson 25 (When to Trust the Output) and Lesson 26 (How to Catch a Hallucination) take the practical workflows that follow from these limits.

What you have, what comes next

You now know what the model’s confidence actually measures, why confident wrongness is structural rather than accidental, and why two re-asks of the same prompt can disagree. You have met the term hallucination here as a conceptual seed; the term gets named, defined, and made useful in Lesson 15.

In Lesson 10, we pull the camera back one more step. The probabilities the model uses to predict the next token were not handed to it by anyone. They were learned, from a procedure called training, which is the central engine of every machine-learning system in the world. Next week’s lesson is what training actually is, in plain English, with no calculus. Once you understand training, the central principle from Lesson 4 (data is the model) will move from a phrase to a piece of machinery you can see working.

If You Want to Dig Deeper

For the technical-paper-but-readable treatment of why neural networks are systematically overconfident, Guo et al. (2017) is the canonical empirical reference and the source of the calibration vocabulary in the field. Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (PMLR 70, pp. 1321–1330). https://doi.org/10.48550/arXiv.1706.04599

For the cleanest treatment of whether asking a model are you sure? works (the careful answer being: partially), Kadavath et al. (2022) is the relevant Anthropic study. Kadavath, S., Conerly, T., Askell, A., Henighan, T., Drain, D., Perez, E., Schiefer, N., Hatfield-Dodds, Z., DasSarma, N., Tran-Johnson, E., Johnston, S., El-Showk, S., Jones, A., Elhage, N., Hume, T., Chen, A., Bai, Y., Bowman, S., Fort, S., … Kaplan, J. (2022). Language models (mostly) know what they know (preprint). arXiv. https://doi.org/10.48550/arXiv.2207.05221

For the broader conceptual essay this lesson sits on top of, Murray Shanahan’s piece in Communications of the ACM remains the cleanest short explainer. Shanahan, M. (2024). Talking about large language models. Communications of the ACM, 67(2), 68–79. https://doi.org/10.48550/arXiv.2212.03551

Author Note:

Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. She corrects all her papers and articles with Grammarly, because even though she has deep thoughts, she has shallow patience for punctuation. She uses Anthropic’s Claude in Research mode for source location and verification on cited factual claims; all interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

References

Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (PMLR 70, pp. 1321–1330). https://doi.org/10.48550/arXiv.1706.04599

Kadavath, S., Conerly, T., Askell, A., Henighan, T., Drain, D., Perez, E., Schiefer, N., Hatfield-Dodds, Z., DasSarma, N., Tran-Johnson, E., Johnston, S., El-Showk, S., Jones, A., Elhage, N., Hume, T., Chen, A., Bai, Y., Bowman, S., Fort, S., … Kaplan, J. (2022). Language models (mostly) know what they know (preprint). arXiv. https://doi.org/10.48550/arXiv.2207.05221

Shanahan, M. (2024). Talking about large language models. Communications of the ACM, 67(2), 68–79. https://doi.org/10.48550/arXiv.2212.03551

Read more

Start with what’s true, because the truest part is the part that the framing depends on you skipping past.

On the night of July 4, 2025, outside the Prairieland ICE Detention Center near Alvarado, Texas, a police lieutenant named Thomas Gross was shot in the neck. It exited his back. He was flown to a hospital, treated, released, and he lived. The man who fired, Benjamin Song, was sentenced this week to 100 years in federal prison. Seven of his co-defendants were sentenced the same day. Together they received 450 years (Department of Justice, 2026).

None of that is a hoax. A jury in the Northern District of Texas heard twelve days of evidence and convicted these people in March (CBS News, 2026). The press release announcing the sentences is a genuine document currently on the United States Department of Justice website, release number 26–685. When you ask, “Is this real?” the answer is yes. A man was shot. People are going to prison for a very long time.

So, I want to be careful, because the easiest way to be wrong about this story is to treat the violence as fake. It’s not fake. Hold onto that. Then read the headline again.

“Leader of Antifa Cell.”

Three words are doing an enormous amount of quiet work. Cell. Leader. And underneath both, the thing neither word says out loud, but both require: an organization. A structure with a top and a bottom, a command and a rank, a membership you can join, and a roster you can be on. That’s the picture the headline paints. That’s the picture I want to take apart, because the shooting is real and the organization is not, and the gap between those two facts is the whole story.

The thing that does not exist

Here is the inconvenient part for anyone selling you the cell.

The last time a sitting FBI director testified under oath about antifa, he said it was not an organization. Christopher Wray, appointed by Donald Trump, told Congress in 2020 that antifa is “not a group or an organization” but “a movement or an ideology.” The Congressional Research Service, the nonpartisan body that exists to tell legislators what is actually true, says the same: there is no national antifa, no headquarters, no leadership, no membership (Congressional Research Service). The historian who wrote the standard book on the subject, Mark Bray, has spent years explaining that the word refers to a loose politics, a tactic, a banner anyone can pick up and put down. He left the country in late 2025 after the death threats started.

You don’t have to take the scholars’ word for it. Take the word of the government’s own witnesses in this trial. One cooperating defendant testified that antifa, as he understood it, was just an idea about being opposed to fascists, not a group anybody belonged to (CBS News, 2026). Several of the people who took plea deals said on the stand that they were surprised when the protest turned violent, because they hadn’t signed up for violence. There was nothing to sign.

The government brought an expert to say otherwise. He holds a bachelor’s degree. He had never testified as an expert before. His work is not peer-reviewed, and he conceded as much. He told a reporter, in plain words, that he gave prosecutors a definition of antifa he believed was accurate, and they used it (Lawfare, 2026). That’s the evidentiary spine of the word “cell” in a federal headline: one man’s preferred definition, repeated back by the office that asked for it.

And “leader”? Benjamin Song shot a police officer, and for that, he will likely die in prison, and a jury decided that. But “leader of a cell” is not a crime he was convicted of. There is no leadership count in the case. It’s a word in a press release, a background allegation that the others “looked to” him, not a thing prosecutors had to prove to twelve people beyond a reasonable doubt. The conviction is real. The org chart is a decoration.

How spray paint becomes terrorism

This is where the machinery gets interesting, and where the can of spray paint comes in.

To call this a terrorist attack, prosecutors used a statute called 18 U.S.C. § 2339A, material support to terrorists. It’s worth knowing what that law does and does not require. The better-known terrorism statute, § 2339B, requires a designated terrorist organization before it can be charged. Section 2339A needs no organization at all (Congressional Research Service; 18 U.S.C. § 2339A). It only needs an underlying federal crime of terrorism to attach the support to. So the “cell” language was never legally necessary for the terrorism charge. It was necessary for something else.

The underlying crime they leaned on, for many of these defendants, was depredation of government property under § 1361. In this case, that means the spray paint. Somebody painted “ICE pig” on a vehicle and turned off a camera. That vandalism, prosecuted as property destruction, became the predicate that turned a protest into terrorism and put fifteen-year exposures on people’s heads. Legal analysts who looked could not find a recent precedent for using property destruction in this way to anchor a material-support-to-terrorism charge (Lawfare, 2026).

So look at what one defendant got. Daniel Sanchez-Estrada was not at the protest. He was nowhere near Lt. Gross. His conduct included concealing records and moving a box of zines. He was sentenced to 30 years. The distance between “moved a box of pamphlets” and “30 years for terrorism” is the distance the word “cell” is built to make you forget.

The timeline gives it away.

If the organization were real, you would expect the terrorism charges to have come first, at the moment of the shooting, when an officer was bleeding in a parking lot.

They did not.

When these people were first charged, the charges were attempted murder and firearms offenses, serious, violent, straightforward crimes. The terrorism language and the “North Texas Antifa Cell” framing were not in the original indictment. They were added later, in superseding indictments returned in November and December of 2025 (Lawfare, 2026).

What happened in between is the thing to keep your eye on. In September 2025, after the killing of Charlie Kirk, President Trump signed an executive order designating antifa a domestic terrorist organization (NPR, 2025). The language in that order, the part about a “militant enterprise” bent on overthrowing the government, is nearly the same language that then appeared in the indictment describing this group (Lawfare, 2026). The protest happened in July. The order came in September. The terrorism framing followed in November. A shooting in the summer was reverse-engineered, in the fall, into the first proof of a designation signed in between.

Legal scholars have been blunt about that designation. No statute lets a president brand a domestic group a terrorist organization. The Brennan Center put it plainly: the administration has no authority to do this, and the purported designation has no legal effect (Brennan Center for Justice, 2025). The State Department can designate foreign organizations. There is no domestic equivalent, because the First Amendment is the domestic equivalent, and it points the other way.

What is the word for

So here is what the framing accomplishes, once you strip the violence away from the decoration.

A real shooting gives a designation that has no legal basis, something to stand on. An executive order signed in September needs a body, and a case filed in July can be made to supply one if you add the right words to the indictment. The administration gets to announce its first antifa terrorism conviction, a template it has promised to repeat. The FBI director vowed this week to dismantle antifa and its funding networks across the country, which is a strange thing to promise about a movement with no networks and no funds, unless the promise is the point.

This is what the “terrorist” label has always been good for. Scholars who study political violence have noted for decades that the word is not neutral. It’s a verdict dressed as a description. It sorts people into the category that forfeits sympathy, and it lets ordinary crimes, the kind we already have laws for, get reclassified as something that justifies extraordinary power. Naming a cell and naming a leader takes a dozen people who knew each other through a Signal chat and fits them into the shape of al-Qaeda, because that shape frightens, and frightened people don’t ask whether the shape is true.

I’ll give the government its strongest case, because it has one. These were not pacifists holding candles. They coordinated via encrypted chats that self-deleted. They brought Faraday bags to kill their phone signals, body armor, eleven firearms, and military trauma kits stocked with tourniquets. A jury looked at all of it and saw planning, not spontaneity, and a man really was shot in the neck. If you want to call that a conspiracy to commit violence, the evidence is there, and I wouldn’t argue with you.

But a conspiracy among a dozen people is not a national terrorist enterprise, and the case knows it. The same jury that convicted Song acquitted four other defendants of the violent charges entirely. One man got 30 years for a box of zines he carried while standing nowhere near the scene. The terrorism enhancement rode into court on a can of spray paint. And years before any of this, in 2018, the FBI’s own Dallas field office had opened an investigation into antifa activity in the very same region, looked hard, and closed it, finding no criminal threat worth pursuing. That file existed. The defense said it wasn’t handed over the way it should have been.

A can of spray paint is now, in the government’s theory, an instrument of terrorism. A box of pamphlets is now worth 30 years. Not because the violence was invented, but because the violence was real enough to hang a much larger and emptier word on, and to hope you would not check whether anything was underneath.

The shooting was real. Check the rest yourself.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

References

Brennan Center for Justice. (2025, October 1). Trump’s version of “domestic terrorism” vs. the First Amendment. https://www.brennancenter.org/our-work/research-reports/trumps-version-domestic-terrorism-vs-first-amendment

CBS News. (2026, June 23). Leader of group convicted in antifa-inspired attack on Texas ICE facility handed 100-year prison sentence. CBS Texas. https://www.cbsnews.com/texas/news/ice-detention-attack-defendants-sentencing-6-23-2026/

Congressional Research Service. Terrorist material support: An overview of 18 U.S.C. § 2339A and § 2339B (R41333). https://www.congress.gov/crs-product/R41333

Department of Justice, Office of Public Affairs. (2026, June 23). Leader of antifa cell, members in North Texas, sentenced to 100 years in prison for terrorist attack on ICE facility (Release №26–685). https://www.justice.gov/opa/pr/leader-antifa-cell-members-north-texas-sentenced-100-years-prison-terrorist-attack-ice

Lawfare. (2026). Lawfare Daily: The trial of the North Texas antifa cell. https://www.lawfaremedia.org/article/lawfare-daily--the-trial-of-the-north-texas-antifa-cell

NPR. (2025, September 19). Trump signs an order calling “antifa” a domestic terrorist group. https://www.npr.org/2025/09/19/nx-s1-5545764/trump-antifa-domestic-terrorist-organization

U.S. Code. 18 U.S.C. § 2339A, Providing material support to terrorists. https://www.govinfo.gov/link/uscode/18/2339A

Watch a chatbot answer you sometime. The text streams in. Word by word, or sometimes a piece at a time, never quite all at once. The stream is not a transcription of a finished answer the system has already composed. The stream is the system, in real time, doing exactly one thing: predicting what comes next.

That one thing is the engine of every generative language model on the market. Not metaphorically, not loosely, but in the actual technical sense of what the program is doing at each step. The job is to guess the next token. Then guess the next one. Then the next. ChatGPT, Claude, Gemini, the autocomplete in Gmail, the half-line of code that pops up in Copilot, the suggestion that appears under what you are typing on your phone: all the same operation, applied at different scales, in different products, to different vocabularies, and with different amounts of context. The mechanism, end to end, is shorter than any of the systems built on top of it.

This lesson opens the engine and walks through the prediction game in plain English. By the end you will know what the model is doing every time you watch it write a sentence. The rest of Part 2 builds out from this one operation.

Read more

I am a transgender woman. In 2009, before I had the word for what I was, a group of people in Washington decided that protecting people like me was part of a hate crimes bill you could give away to make it move faster.

They were wrong about who would let them. Nancy Pelosi, then Speaker of the House, has now put their offer on the record. “People told me, ‘You can pass this in a minute if you take out trans,’” she told the Washington Blade this June. “I said, ‘I won’t pass it in 100 years because I’m not ever taking out trans.’ We passed it with trans protections included.”

The headlines that followed called it good news, and it is. Read the sentence again, though, and notice what it assumes. Before anyone could refuse to take trans people out, someone had to ask. The story worth telling is not the refusal. It is the request.

So I want to ask the question that the celebration skips. Who wanted trans people cut from a law about violence, and what did they think they would get for it?

What the law actually says

Start with the part that is not in dispute. The Matthew Shepard and James Byrd Jr. Hate Crimes Prevention Act became law on October 28, 2009. The statute it created, 18 U.S.C. § 249, makes it a federal crime to injure a person for their actual or perceived “religion, national origin, gender, sexual orientation, gender identity, or disability.” Gender identity is in the text. Pelosi’s “we passed it with trans protections included” is not a memory she is polishing. It is the law of the United States.

Which means the pressure she describes was a push to write a smaller law. One that would have said the federal government takes violence against a gay man seriously and has nothing to say about violence against a trans woman. That smaller law had supporters in both parties. Some of them were supposed to be on my side.

Inside the tent

The pressure on Pelosi was neither hypothetical nor new. Two years earlier, the same maneuver had already run in her own chamber, with her cooperation.

In the fall of 2007, Representative Barney Frank took the Employment Non-Discrimination Act, a bill barring employers from firing workers for being gay or transgender, and split it. He introduced a version covering sexual orientation that dropped gender identity, arguing the votes for an inclusive bill were not there. The gay-only version passed the House 235-184 on November 7, 2007. Pelosi brought it to the floor.

The reaction told everyone what had happened. More than 300 LGBTQ organizations formed a coalition called United ENDA to fight the split, refusing a bill that protected part of the community by leaving the rest out. Frank’s defense then is the same one offered for every version of this trade. Years later, he put it plainly: the transgender community, he told Georgia Voice, had a “mistaken view that if Nancy Pelosi waved a magic wand, transgender would be included,” and the votes simply were not there.

To her credit, Pelosi spent that fight trying to undo the damage, backing an effort by Representative Tammy Baldwin to restore gender identity. That effort was lost. The lesson she seems to have taken from it shows up two years later, when the people doing the vote math came back with the same offer on hate crimes, and she refused it. That is the inside of the tent: trans people as the part you sacrifice for the win, and one leader who decided, the second time, not to.

Outside the tent

The reason the votes were short is that they have names.

In public, the opposition wore a suit. The bill was unnecessary, its opponents said, a federal intrusion on work the states already did. When the House first passed an inclusive version in 2007, the Bush White House issued a veto threat the same day, objecting that there was “no persuasive demonstration of any need to federalize” such a large range of violent crime. Representative Tom Cole called it a “thought crimes bill” that would divide Americans “into different classes of people subject to different protections under the law.” Equal protection, they argued, meant no special protection. It was a respectable argument, and it was a costume.

Underneath it was something else. On the House floor on April 29, 2009, with Matthew Shepard’s mother watching from the gallery, Representative Virginia Foxx called the murder that gave the bill its name “a hoax,” insisting the young man “was killed in the commission of robbery. It wasn’t because he was gay.” She later called “hoax” a poor choice of words. Judy Shepard, who had been sitting a few feet away, said the apology was “for semantics but not her sentiment, her insensitivity or her ignorance.”

The same week, Representative Steve King told Fox News the bill would create “special protection for pedophiles,” a claim PolitiFact rated false, noting that federal law already defines sexual orientation as heterosexuality or homosexuality and excludes pedophilia. Representative Louie Gohmert went further on the floor, warning that protecting people from violence based on orientation would mean striking the laws against sex with “animals,” “corpses,” and “children,” and would walk the country toward 1930s Germany.

None of that was about federalism. You do not reach for bestiality and Nazi Germany to make a point about which level of government prosecutes assault. The procedural objection was the public reason. The contempt was the real one. And the contempt is what set the vote count, which the people inside the tent then cited when they told Pelosi to make it easier on herself.

That is the turn the good news hides. Expediency and hate were never two stories. The expedience was the hate, run through an adding machine. When an advisor said, “You can pass this in a minute if you take out trans,” the minute he was promising had been bought by Foxx and King and Gohmert, by every member whose vote came free only if my existence stayed outside the law’s protection.

One fact disposes of the stated reason entirely. The bill, the opponents called a gag on pastors, contained an evidentiary rule barring a defendant’s speech and associations from being used against him at trial except where they related directly to the crime. The claim that it would muzzle preachers was false on its face, as a matter of the statute its critics were reading. They made it anyway, which tells you the argument was never the point.

Why

So, the question my editor would say I have been circling. Why? What did the people leaning on Pelosi actually want?

I cannot read what was in their hearts, and I am not going to pretend the record lets me. What the record shows is a goal more durable than any single vote. They wanted trans people to stay the negotiable category. Not protected, not even singled out for harm, just available, the line you can always offer up when a deal needs grease. To cut us from a law about violence is to enter a quiet verdict that violence against us counts for less. That verdict, not any one congressman’s disgust, is the thing they were defending.

It is the same verdict being entered right now.

The same minute, again

In January 2025, the President signed an executive order titled “Defending Women from Gender Ideology Extremism and Restoring Biological Truth to the Federal Government.” Read the title slowly. It seats the word “extremism” beside an identity in the heading of a federal directive. Gregory Stanton, who developed the framework the field uses to track how this escalates, lists classification, symbolization, and dehumanization as early stages. An official document that files a minority under “extremism” is not being shy about which stage it performs.

The line is being cut in real time. In June 2026, Lambda Legal sued the Department of Health and Human Services over new rules barring Ryan White HIV program funds from being used to acknowledge or treat transgender patients, pulling trans people out of a program that has kept low-income people with HIV alive for three decades. Days earlier, a federal judge blocked the Bureau of Prisons from transferring fourteen transgender women into men’s prisons, in a system where, by the government’s own data, trans people are sexually assaulted at nearly ten times the rate of other prisoners. Pelosi, who retires in 2027, named the pattern herself last year. “The trans issue is what they have glommed onto,” she told The Advocate.

Take out trans, and the program runs cheaper. Take out trans, and the prison policy gets simpler. Take out trans, and the bill passes in a minute. It is the same sentence every time, and trans people are always the words being removed.

What it cost

I came out at fifty-three. I do not get those years back, and most days I have made my peace with that. What I have not made peace with is watching, in real time, an argument I thought my country had finished.

In 2023, the FBI recorded 492 hate crimes based on gender identity, the highest annual total since it added the category in 2013. That is the number the Matthew Shepard Act exists to count. The law that almost got me cut out is now the law measuring how often people like me are attacked, and the count climbs as the same coalition works to pull the protections back out. The people who told Pelosi to drop us were not wrong that it would have been easier. They were wrong about what easier was worth.

She gave them a hundred years. The people now reaching for the same minute should understand that some of us intend to make it cost at least that long.

I am still here. We are the line they keep trying to delete, and we are still in the text.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

In 1999, a Texas farmer named Frank Rhea Cromwell signed away 87.97 acres for $10, and one sentence at the top of the deed said why: the land was “to be held in trust for future use as parkland by Williamson County, Texas.” In 2025, fifty-three of those acres sold for $10 million to a data center developer. And this past October, a judge ruled that the neighbors to whom the park was promised had no standing to object.

Taylor, Texas, is a small city northeast of Austin. Pamela Griffin grew up on the south side, in a three-block pocket that was one of the first places in Taylor where Black and Hispanic families could buy residential lots after the 1968 Fair Housing Act, Richard Stone reported for the Austin Free Press. About 35 families live there now, most on land the original families still hold. Griffin’s grandmother bought on the outskirts in the years when Black residents could not buy inside the city limits at all. Her father bought a vacant lot so Griffin and her ten siblings would have somewhere to play, and that lot backed up against Frank Cromwell’s farm.

Cromwell waved at the kids from his tractor. He let them play baseball in his field and fish for minnows. And he said the quiet part out loud to Griffin’s father: “If something happens to me, I’m going to give this land to y’all for a park.”

“We did not know that man kept his word,” Griffin told KUT’s Kailey Hunt in September 2025. “He knew we needed something, and he was willing to give it to us.”

One sentence at the top of a deed

He kept it in writing. On July 7, 1999, the Cromwell family deeded the 87.97 acres to the Texas Parks and Recreation Foundation for $10, with the purpose stated in a single line: “to be held in trust for future use as parkland by Williamson County, Texas.” Matthew Gault reviewed a copy of that deed for 404 Media this June. The sentence is there.

What happened next is a chain of paperwork, and the sentence did not survive it. The chain is laid out in Gault’s investigation and Stone’s reporting:

1. In 2003, the foundation handed the land to the Williamson County Park Foundation, which handed it to the City of Taylor a month later.

2. In 2005, Taylor zoned the tract industrial. No park was ever built.

3. In 2008, the city sold the land to its own economic development corporation for $15,000.

4. In 2023, a new land development code re-labeled the area “Employment Center,” a designation under which a data center is already a permitted use.

5. In 2025, the Taylor Economic Development Corporation sold 53 acres to Blueprint Data Centers for $10 million, and 27 adjoining acres to a steel company.

The parkland sentence was omitted from the later transfers, so the development corporation held the land with no restriction visible on its face. A $10 gift, held in trust for children, converted into $10 million of sellable ground in five clerical moves. Nobody had to repeal the promise. They just stopped copying it forward.

What’s coming instead

Blueprint Data Centers is an Austin developer that bought the ground through a parent entity, NCP Travis BPP Project LLC, and took a strategic investment in April 2025 from Northampton Capital Partners, an asset manager with about $1.1 billion under management. The plan, per the company’s own announcements and the Austin American-Statesman’s Claire Osborn: a 135,000-square-foot data center built in three phases, up to $1 billion in capital over its first decade, and 30 megawatts of committed electricity scaling to 60. It will sit about 500 feet from Pamela Griffin’s home.

The Taylor City Council approved the tax deal on August 8, 2024: a 50% property tax rebate for ten years on each of the three construction phases, plus half the local use tax on construction materials handed back, Osborn reported. The city projects roughly $30 million in new revenue over the decade. And the company’s binding jobs commitment, in the city’s own news release: a minimum of five.

Residents found out late. Carrie D’Anna spotted the project on a council agenda and started a Facebook group and a flyer run; Griffin walked her three blocks handing flyers out, Hunt reported. Neighbors spoke at hearings in June and July of 2025, and the council approved the project’s site plan unanimously, granting a 30-day delay rather than the 90-day moratorium the neighborhood asked for. Mayor Dwayne Ariola has described the project as “a concrete room with computer servers in it,” per Gault. He offered that as reassurance.

The city’s own FAQ asks, “Can the City just say no to data centers?” Its answer, as Justin Caffier reported at Gizmodo: “In short, no.” The zoning was already in place. Taylor’s development arm collected the $10 million; Taylor’s FAQ pleads helplessness. Both statements are true, and that is the problem.

No standing

Pamela Griffin, three of her siblings, and a neighbor named Polly Randle hired a Taylor lawyer, Chris Osborn, and sued to stop all commercial development on the site, the deed at the center of their case. Blueprint’s lawyers argued the deed contains no enforceable restriction at all, and a real-estate appraiser testified for the defense that deed restrictions, in his experience, come as specific clauses in the body of a document, not a single line at the top, Edie Zuvanich reported for the Taylor Press. Even the 15-acre buffer the city kept between the site and the houses is not planned as a park; the city’s comprehensive plan marks it for commercial- and residential-type development, Taylor’s assistant city manager testified.

On October 8, 2025, Judge Ryan Larson of the 395th District Court in Georgetown denied the injunction and dismissed the case. Not on the deed. On standing: the ruling held that the neighbors are not parties entitled to enforce it. “We were disappointed that he denied the injunction, of course, but we were really surprised that he also dismissed the case,” D’Anna told Zuvanich. The family has appealed to the Texas Third Court of Appeals, where the case is pending as of this writing.

Here is the part worth slowing down for. Texas law does protect parkland. Chapter 26 of the Parks and Wildlife Code requires three things before land “dedicated and used as a park” can be taken for another use: notice, a public hearing, and a finding that there is “no feasible and prudent alternative.” Used. Taylor never built the park, so the strongest shield in Texas law for park ground may never attach to ground that spent 26 years waiting to become one. And charitable gifts in Texas are enforced, for the most part, by the attorney general, not by the neighbors a gift was meant for. The promise was broken before it was kept, and the not-keeping is what put it past the neighbors’ legal reach.

Read that paragraph again. The defense never claims the promise was not made. The defense claims nobody left in that courtroom is allowed to care.

The arithmetic Texas is doing

Taylor is not an outlier; it is the pattern. Texas now has more than 300 data centers operating and, per an Aterio analysis that the Texas Tribune’s Paul Cobler and Apurva Mahajan reported in April, 142 under construction, one more than Virginia. The state comptroller estimates the sales-tax exemption for data centers will cost Texas $3.2 billion over the next two years. Organized opposition has surfaced in San Marcos, Amarillo, College Station, Waco, and Harlingen. The buildings keep coming anyway, and they go where the ground is cheap and the resistance is presumed to be manageable.

Ground gets cheap, in part, by sitting next to people whose objections have historically been easy to discount. The 53 acres in this story were not pried away from a country club.

Where I stand in this

I should be honest about my seat at this table. I am a doctoral student in artificial intelligence. The models I study have to live somewhere, in buildings exactly like the one rising near Pamela Griffin’s porch, and my field’s appetite is the demand side of this story. I typed this essay on the supply chain that ate her park. That is not a reason to quit working in AI. It is a reason to refuse the comfortable fiction that the buildings land where they land by accident.

The deed is still on file in Williamson County. One sentence at the top, $10 at the bottom, and 26 years of paperwork insisting the sentence stopped counting somewhere between the courthouse and the closing table. Frank Cromwell watched other people’s children play on his land and wrote them a promise into the one kind of document Texas claims to hold sacred. He kept his word.

The appeal will decide whether anyone else has to.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

References

Blueprint Data Centers. (2025, August 13). Blueprint Data Centers delivers accelerated digital infrastructure readiness with strategic Austin facilities, energized by 2026 [Press release]. Telecom Ramblings Newswire.

Business Wire. (2025, April 14). Northampton Capital Partners invests in Blueprint Data Centers [Press release].

Caffier, J. (2026, June 8). A farmer donated land for a public park and the city sold it to a data center developer for $10 million. Gizmodo.

Cobler, P., & Mahajan, A. (2026, April 8). Texas losing a billion dollars a year on data center tax break. The Texas Tribune.

Gault, M. (2026, June 8). A farmer donated land to turn into a park. The city is building a massive data center instead. 404 Media.

Hunt, K. (2025, September 26). It was supposed to be a park. Instead, this property in Taylor is being turned into a data center. KUT.

Osborn, C. (2024, August). BPP Projects plans to build new $225 million data center in Taylor. Austin American-Statesman.

Stone, R. (2026, May 18). Minority report: Taylor data center plan raises community ire and questions of equity. Austin Free Press.

Texas Parks and Wildlife Code, ch. 26. Protection of public parks and recreational lands. Texas Legislature.

Zuvanich, E. (2025, October 1). Blueprint Data Centers hearing kicks off. Taylor Press.

Zuvanich, E. (2025, October 11). Judge rules in favor of data center. Taylor Press.

A psych-ward comedy bit and the question the Second Coming can’t answer.

We Killed Him Once.

A friend of mine sends me stand-up clips the way other people send recipes.

The last one was Dan Docimo, taped at a club in Chicago, and it’s been rattling around my head for weeks. He says he was talking to a friend who works in a psych ward. She told him how often patients come in believing they are Jesus.

Not once in a while. Constantly. All over the world, ward after ward, there’s usually a guy or two who are certain they are Christ.

Docimo does the math out loud. If the number is that big, he figures, “one of them is probably just Jesus, because that was, like, a big thing for him, is that he was going to come back.”

Then he springs the trap.

Anyone who walks outside today and says he is the returned Christ gets walked straight into a locked ward. “There is actually not a step in the process where they make sure you aren’t Jesus,” he says. “They really just assume that you’re wrong.”

So he wonders how many times Jesus has already come back and died in an institution, anonymous, on a vinyl mattress. His guess is dozens. “I bet Jesus has been administered more Thorazine than anyone in human history.”

I laughed.

Then I stopped because the joke rests on a real problem, and that problem doesn’t have a comfortable answer.

Start with the part you can look up, the part Docimo got right. The delusion is real and not rare.

A psychologist named Milton Rokeach spent two years on exactly this. In 1959, he gathered three men at a Michigan state hospital, each convinced he was Jesus Christ, and put them in a room together to see what would happen when one unshakable identity met two more just like it (Rokeach, 1964).

They didn’t back down. Each man simply explained the other two away.

Rokeach later wrote that he had had no right, even in the name of science, to play God in those men’s lives. The apology reads like its own small sermon.

The numbers back the frequency. Among people living with schizophrenia, religious grandiose delusions are common, and the content runs to the obvious headliners: a saint, a prophet, the devil, Jesus himself (Murray, Cunningham, & Price, 2012).

There is even a version that visits the supposedly healthy. Doctors at a Jerusalem psychiatric center described tourists with no history at all who arrive in the city and come apart, a few of them wrapping themselves in hotel bedsheets and heading out to preach (Bar-El et al., 2000).

Honesty requires a footnote.

Other psychiatrists pushed back hard, arguing that nearly everyone affected was already ill before they landed (Kalian & Witztum, 2000), and the condition appears in no diagnostic manual.

The fight doesn’t soften the picture. It sharpens it. Even the experts can’t draw a clean line between the man touched by God and the man whose brain has turned on him.

And the drug is real. Docimo names Thorazine, and Thorazine is not a punchline he invented.

Chlorpromazine, sold in America under that brand from 1954, was the first antipsychotic (Ban, 2007). It did to the asylums what nothing before it had managed. The population of American state mental hospitals stood north of half a million in the mid-1950s, then fell, year after year, as the drug spread and the wards emptied (American Chemical Society, 2005).

So, the instrument in the joke is the actual instrument. If Christ showed up tomorrow and said so on a street corner, the medication to quiet him has been sitting on a shelf for seventy years.

Here is where the comedian’s bit turns into a question for the people in the pews. I count myself one of them, on my better Sundays.

I run the sound booth at a country church outside Sioux Falls. I’m a skeptical agnostic Lutheran, which means I take the prophets seriously and the prophecy-chart salesmen not at all.

So let me ask the thing the joke keeps circling back to. What did we actually say the return would look like?

The tradition I was raised in answers loudly. The Augsburg Confession, the document that spells out what Lutherans confess, says that Christ will return visibly at the end of everything to judge the living and the dead (Melanchthon, 1530/2024).

The scriptures behind it promise the same scale. Matthew compares the coming of lightning that flashes across the sky with the Son of Man arriving on the clouds in glory (Matthew 24:27, 30). After the ascension, two figures tell the disciples he will come back the same way they just watched him go (Acts 1:11).

Visible. Glorious. Unmistakable.

Nobody could confuse that with a man in a paper gown.

That’s the expectation. Now set it next to the record of the first time.

Because the same texts that promise the blazing return also tell us, plainly, how the man went over when he was actually here.

Isaiah, in the passage the church reads as his portrait, says he was despised and rejected, a man of sorrows, from whom people hid their faces (Isaiah 53:3).

Mark says his own family went out to grab him, convinced he was out of his mind (Mark 3:21).

John records the crowd’s verdict in words a triage nurse would recognize: “He has a demon, and is insane; why listen to him?” (John 10:20).

In the first century, they called it demonic possession. We have a different word now, a different building, too, and a drug to go with it.

So how would you tell the difference?

Read the two pictures side by side. The return we advertise is lightning across the sky. The reception we actually gave him was an insanity charge, and a family was trying to walk him home before he embarrassed them.

That’s why the joke lands. It forces together the two things the doctrine keeps in separate rooms.

If the real Christ came back unannounced and unarmed, no army, no clouds, just a man insisting on who he was, every system we’ve built would read him as a patient. The diagnosis would be sincere. The paperwork would be in order. And there’s no step in the process, as the comedian says, where anybody checks.

Dostoevsky saw this coming a century and a half ago.

In the parable his characters tell, Christ comes back to Seville during the Inquisition. The people know him instantly. The Grand Inquisitor has him arrested by nightfall (Dostoevsky, 1880/1990).

The old churchman visits the cell and explains, almost kindly, that the church has finished its work and no longer needs him. That he’s coming back now would only get in the way.

The institution built in his name turns out to have no use for the man himself.

He listens. He says nothing. He kisses the old man and is let out into the dark.

I’m not telling you any of this to mock anyone’s faith.

The dig is at the institution, not the believer, and the difference matters to me because I’m still in the building on Sunday mornings. My skepticism has never pointed at God. It points at what we keep doing with God: the boxes we build and label and then defend with whatever the century puts in our hands. A sword in one era. A syringe in the next.

The hope that the world gets set right someday is not the part I’m questioning.

The machinery we’ve assembled to recognize the one who would set things right is the part that should worry anyone who reads the Gospels closely.

Docimo has one more move, and it’s the one I keep coming back to.

He asks the crowd if they’re excited about Jesus coming back. A few of them cheer. “We’re gonna whoop his ass,” he says. “It’s gonna be a good rematch. I think we’ll get him again. We didn’t even have guns last time.”

It’s a filthy joke. It is also the most honest reading of the Gospels I’ve ever heard from a comedy stage.

We did get him the first time. The crowd waving palms on Sunday was calling for a cross by Friday. The texts don’t hide that. They put it at the center.

So I’ll leave the question where the comedian left it. On the vinyl mattress. In the locked ward. With the man saying the one thing nobody in the building is built to hear.

We already killed him once.

We aren’t the people who would know him if he came back.

We’re the people who assume you’re wrong.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

Open a fresh chat window in ChatGPT or Claude and ask, “How many r’s are in the word strawberry?” There is a documented chance you will be told two. There is a smaller documented chance you will be told four. There is some chance you will be told three, which is the right answer. Then ask the same question again in a new conversation. The answer may change. The model is not stupid, and it has not lost interest. It is doing its job exactly as designed, and its job does not involve looking at letters.

Welcome to Part 2. Last week we closed Part 1 with the question of whether the system understands what it is doing. This week we open the engine. The first thing to notice, the smallest thing inside the machine, is the unit it actually reads. That unit is not a word. It is not a letter. It is something called a token, and almost every weird thing about how language models behave starts here.

Read more

HB1232 is frightening on its own text. Inflating it costs us.

D. Denise Dianaty is right that North Carolina House Bill 1232 should frighten you. She is wrong about how to make you afraid of it.

Her piece, “NC Republicans to Legalize Murder,” reads the bill correctly at the center and then keeps reaching past what the record will hold. The reaching is the problem. The bill itself is real, and parts of it are worse than most readers will guess. What undoes the piece is that every claim past the evidence hands the other side a reason to throw out the claims that rest on it. I want to separate the two, since the first set is the case, and the second set is what gets the case dismissed.

What the bill actually says

Read it in its own words, on the General Assembly’s own site. HB1232 would add a section to Article I of the state constitution declaring it “a matter of indisputable scientific fact that a distinct and separate human life begins at the moment of fertilization.” It would grant that life the standing of “a person entitled to the protection of the laws of this State from the moment of fertilization until the moment of natural death.”

Then it does the thing the headlines are about. The bill says: “Any person who willfully seeks to destroy the life of another person, by any means, at any stage of life, or succeeds in doing so, shall be held accountable for attempted murder or for first degree murder, respectively.” And one sentence later: “Any person has the right to defend his or her own life or the life of another person, even by the use of deadly force if necessary, from willful destruction by another person.”

That is the bill. Not a paraphrase, not a hostile reading. A one-page amendment that defines a fertilized egg as a person, makes destroying that person first-degree murder, and grants a private right to use deadly force in defense of “another person.” It carries no exception for the pregnant woman, none for an ectopic pregnancy that will kill her, and it never mentions contraception or IVF at all. PolitiFact read the same text in June 2026 and described it the same way: a proposal to define life at fertilization and grant the right to defend the unborn, potentially through deadly force. Quote that, link the PDF, and you have done the reader a service. The text frightens on its own.

Where the article starts reaching

The headline says the bill “gives explicit permission to murder women & doctors.” The deadly-force language is real, so the conclusion is not crazy. It’s a conclusion, though, built by stacking the personhood clause on the deadly-force clause and noticing there is nothing in between to stop someone from drawing the obvious line. “Could be read to authorize private violence” is a sentence I can defend to a hostile reader. “Gives explicit permission to murder” is a sentence that the reader gets to call a lie, since the bill does not say it in those words. The gap between those two sentences is exactly where you lose the person you most need to reach.

Same with the contraception claim. Dianaty writes that the amendment “will criminalize IUDs and morning-after contraceptives.” The bill says nothing about either. The worry is real and worth raising, since it turns on whether a court treats fertilization, not implantation, as the line, and the mainstream clinical definition of pregnancy starts at implantation. PolitiFact quoted a public-health expert making exactly that point. “Could be used to reach IUDs” is true. “Will criminalize IUDs” is a prediction dressed as a fact.

And the timing. The piece tells you the amendment will be on the ballot “this November.” A North Carolina constitutional amendment reaches the ballot only after it clears a three-fifths supermajority in both chambers, and this one had not passed a single committee. House Speaker Destin Hall, who controls what gets a vote, said the bill “doesn’t have a chance of moving at all” and called it not serious. Leaving that out makes a stalled message bill sound like a freight train. The honest version is scarier in a different way: this is a trial balloon, filed on purpose, that fails this year and comes back in better clothes. You don’t have to inflate the timeline to make that land.

The Iran comparison is the real problem.

The article opens by calling the bill “Islamic Republic-style legalization of atrocities,” then runs a list of things it says Ayatollah Khomeini legalized after 1979. Dianaty names the risk herself: many readers will think it is outrageous hyperbole. Then she runs at it anyway, and the list does not survive a check.

Then there is the line about an infant, set in quotation marks. Quotation marks are a promise that those were the words. Go looking for the words in Khomeini’s own text, though, and the promise comes apart. The lurid version traces to advocacy sites and unverified translations, not to the source itself. The term his real rulings on child marriage turn on “thighing, ”which does not mean anal contact. The exact quoted sentence is hard to locate in the original text. His actual rulings are ugly enough without a worse one bolted on, and a single invented quotation lets a hostile reader doubt the rest, including the true and damning items below.

The list says Khomeini “legalized sodomizing boys.” Iran does the opposite. Same-sex intercourse is a capital crime there, and the state has executed people for it, as Human Rights Watch has documented for twenty years. Whatever point about child abuse was intended, “legalized” is backward.

The list says Iran “banned and criminalized vasectomies and tubal ligations” and pins it to Khomeini since 1979. For that era, it is the reverse. Iran built one of the developing world’s most admired family-planning programs, free contraception and subsidized vasectomies included, after Khomeini ruled that contraception was not against Islamic law. The crackdown came later, under Khamenei, who, from 2012, began stripping family-planning services and called the old policy a mistake. The claim gets the who and the when both wrong.

What stings is that parts of the list are true. Iran did lower the legal marriage age for girls to nine after the revolution, raised it to thirteen in 2002, with marriage younger still possible by a father’s and a judge’s leave. Temporary marriage is real in Iranian law and works as male-advantaged sexual access. Those facts are documented and damning, and they sit in the same list as an unverifiable quote, an inversion, and a reversed clock, so the skeptic discounts them too.

Here is the paragraph with corrections. Grant the whole list as true, and the comparison still fails. HB1232 is a one-page bill that the sponsor’s own party leadership has ruled out, in a state with courts, a press, and an election coming in November. The Iranian items describe enforced law and documented atrocity under a theocracy with none of those checks. Putting them on the same shelf is the move that earns the “hysterical” label, and it spends credibility the cause cannot spare.

About the sponsor

Dianaty spends a paragraph on the sponsor’s face and on his being a “Yankee who couldn’t cut it in the liberal North.” None of that is an argument. It is the cleanest gift you can give a reader looking for an excuse to call the piece a hit job and close the tab. It’s unnecessary too, since the sponsor’s record is right there, and the record is the point.

His name is Keith Kidwell, a Republican from Beaufort County who chairs the state House Freedom Caucus. He has filed near-total abortion bans before. His name has sat on the leaked Oath Keepers membership roster since 2012, with an entry indicating at least one donation, as Raw Story first reported. State it with the caveat the record itself carries: a name on a leaked donor roster is a documented financial tie, not proof of present, active membership. That caveat is not a hedge. It’s the thing that makes the rest of the sentence stick when someone tries to knock it down.

Why I care about getting this right

I am a transgender woman in South Dakota, and I write about bills like this one, the kind that name people like me as a threat in the heading. I know the machinery by heart. You define a category, you write it into law, you attach a penalty. The fetal-personhood bills run the same machinery, pointed at women instead of at me.

So I do not want it softened or inflated. Both do the same thing to the one thing I have when I tell you a bill is real: whether you believe me. The right’s oldest move is to call us hysterical. Every time our claims outrun our sources, we do that work for them, and the next true alarm lands on a reader who has already learned to discount us.

Quote the bill exactly. Link the PDF. That is the whole job.

The case is good. Don’t be the reason it gets dismissed.

NC Republicans to Legalize Murder

Proposed amendment to the state’s constitution gives explicit permission to murder women & doctors

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

REFERENCES

Guttmacher Institute. (1999). Fertility, contraceptive use and family planning program activity in the Islamic Republic of Iran. International Family Planning Perspectives. https://www.guttmacher.org/journals/ipsrh/1999/06/fertility-contraceptive-use-and-family-planning-program-activity-islamic

Haeri, S. (2014). Law of desire: Temporary marriage in Shi’i Iran (Rev. ed.). Syracuse University Press. https://press.syr.edu/supressbooks/552/law-of-desire/

Human Rights Watch. (2005, November 21). Iran: Two more executions for homosexual conduct. https://www.hrw.org/news/2005/11/21/iran-two-more-executions-homosexual-conduct

North Carolina General Assembly. (2026). House Bill 1232, Const. amend./life at fertilization (first edition). https://www.ncleg.gov/Sessions/2025/Bills/House/PDF/H1232v1.pdf

North Carolina General Assembly. (2026). House Bill 1232 (2025–2026 session): Bill information and history. https://www.ncleg.gov/BillLookup/2025/H1232

PolitiFact. (2026, June 11). Would North Carolina bill allow the ‘murder’ of birth control users? Vague bill likely to stall. https://www.politifact.com/article/2026/jun/11/would-north-carolina-bill-allow-the-murder-of-birt/

Raw Story. (2021). NC GOP lawmaker on Oath Keepers roster says it’s none of the public’s business. https://www.rawstory.com/keith-kidwell-nc-house/

Stimson Center. (2012). Iran: Taking aim at low fertility and women’s mobility. https://www.stimson.org/2012/iran-taking-aim-low-fertility-and-womens-mobility/

United States Institute of Peace. (2020, December 8). Part 3: Iranian laws on women. The Iran Primer. https://iranprimer.usip.org/blog/2020/dec/08/part-3-iranian-laws-women

A trans woman in north London wants to swim in a pond. The machinery now attached to that wish includes a public consultation with tens of thousands of responses, a Supreme Court judgment, a draft code of practice running to over 300 pages, and a court date this autumn. None of that is about the pond.

The gap between those two things is the whole story. The act is small. A person changes in a hut, walks down to the water, and gets in. The structure built around the act is enormous and keeps getting bigger, and the people it is built around are tired in a way that gets misread as weakness, when it is closer to the fatigue of a second job nobody applied for.

What the vote actually did

On 4 June 2026, the City of London Corporation agreed a future access policy keeping the Kenwood Ladies’ Pond and the Highgate Men’s Pond trans-inclusive (City of London Corporation, 2026). Trans women keep their place at the ladies’ pond, trans men at the men’s, and the mixed pond stays open to everyone. The Kenwood Ladies’ Pond is the only women-only wild-swimming spot in the country (City AM, 2024). It opened to the public in the mid-1920s (Guildhall Historical Association, 2016) and has been fought over by the women who swim there for nearly as long, most recently in 2024, when the pond’s own association voted down a motion to bar trans women (PinkNews, 2024).

The vote followed a two-month consultation that ran from late September to late November 2025 and drew more than 38,000 responses, with 86 percent supporting the arrangement already in place (City of London Corporation, 2025). The Corporation’s policy chairman was careful to say the consultation “was never a referendum,” then treated a five-to-one result as the clear instruction it plainly was, the same decision funded upgrades to changing rooms, showers, and toilets, with more private cubicles. Reporting put the package at up to £1.08 million, with final funding due before members in July 2026; the Corporation’s own announcement gave no figure.

So far, this reads like a win, and it is one. A community was asked, the community answered, and the answer held.

The part that did not make the bulletins

Here is the part that did not lead the morning coverage. In March 2026, the Court of Appeal granted the gender-critical group Sex Matters permission to bring a judicial review of the ponds policy, and the substantive hearing is expected this autumn (Local Government Lawyer, 2026). An earlier challenge had already been refused. This is the second pass at the same target.

Take the arithmetic in for a moment, then keep going. The Corporation will now spend public money defending, in open court, a policy that 86 percent of the people it consulted already endorsed. The Trans+ Solidarity Alliance described the position bluntly: an inclusive choice being made to defend itself against what the group called “well-funded lawfare,” with the cost falling on the public purse until the government clarifies its own equality law (PinkNews, 2026). The vote did not end the fight. It scheduled the next round.

It is fair to raise the obvious objection so that I will raise it myself. Critics argue that the 86 percent is soft, that an open consultation invites flooding and tilts away from the regular swimmers it should reflect. The Corporation openly conceded the risk rather than hiding it. It tested the headline number against independent focus groups of regular pond users, who reached the same conclusion (City of London Corporation, 2025). Grant the objection its full weight anyway. The honest answer to a consultation you believe was skewed is a better consultation, or a heavier weighting of the people who actually swim there. The answer on offer is a lawsuit. You do not sue a survey for being unrepresentative. You sue it for returning a result you cannot live with.

One pond, wired into everything

None of this happens in a vacuum. In April 2025, the UK Supreme Court ruled in For Women Scotland Ltd v The Scottish Ministers that the words “woman” and “sex” in the Equality Act 2010 refer to biological sex (UKSC, 2025). The Court was at pains to say the judgment should not be read as the triumph of one group over another, and that trans people keep their protection under the characteristic of gender reassignment. That nuance survived about as long as it took to act on the ruling. Within roughly a week, the equality regulator issued interim guidance on single-sex toilets, washing facilities, and changing rooms (Stonewall, 2025). By May 2026, a draft statutory code of more than three hundred pages had been laid before Parliament.

Consider what a single swimming pond is now wired into:

• a national consultation with tens of thousands of responses,

• a Supreme Court judgment redefining a word,

• a regulator’s interim guidance,

• a three-hundred-page draft code,

• a refused legal challenge,

• a granted judicial review, and

• a hearing not yet held.

A pond. To settle who may swim in it.

This is where the difference between a regulatory question and a banishment question matters, since the two get dressed in the same clothes. A regulatory question has an answer that can be moved by evidence: what hormone threshold, what waiting period, what facility design. A banishment question has only one acceptable answer, and it reverse-engineers whatever process is needed to reach it. When a policy backed five-to-one in consultation still has to run a gauntlet of appeals, the tell is that the process is not searching for the right answer. It already has the answer it wants, and it is willing to litigate until it gets it.

Notice, too, what “inclusive” has been narrowed to mean on the ground. The Corporation has said staff will not check identification or carry out physical assessments, and will rely on judgment and discretion at the water’s edge (PinkNews, 2026). That is humane, and it is the catch. A right that rests on a lifeguard’s discretion, renewed each season and contested in court each year, is not quite a right. It is a permission, and permissions get withdrawn. Living within a permission, never certain which season it will be revoked, is a tax in itself, paid daily in low-grade vigilance by people who only wanted to swim.

Attrition wearing the mask of process

This is the part that sounds like paranoia when trans people describe it, so let me describe it as plainly as I can. The point of a campaign like this is not any single ruling. The point is the calendar.

Another consultation opens. Another code goes out for comment. Another hearing gets a date. Another sign goes up at another pond, another bathroom, another changing room, spelling out who counts and who has to explain themselves at the door. You do not have to win every case to wear down a population. You only have to keep the cases coming, so that defending an afternoon swim becomes a standing obligation, renewed every season, with no version of the future in which it is simply over.

That is the strategy, and it works precisely by not looking like one. It looks like diligence. It looks like consultation, judicial review, and statutory codes, which are the ordinary furniture of a law-governed country. Turned on a single group and kept running indefinitely, that same furniture becomes a method of exhaustion. Attrition wears the mask of process.

Why this is personal, and why that is the point

I should say where I stand. I am not a neutral party. I am a transgender woman. I am, plainly, a woman too, and I have spent years being told those are the same sentence or opposite ones, depending on who needs me to be a threat that week. I live in South Dakota, about as far from a Hampstead bathing pond as a person can get and still be on a map. It does not matter. The grid is transatlantic. The sign at the ladies’ pond in London and the executive order back home that set the word “extremism” next to my identity in its own title are the same instrument, tuned to the same note (Executive Order 14168, 2025).

I came out at fifty-three. I am sixty now. What that sentence skips is the autumn of 2019, when I pointed my car at three grain trucks on a South Dakota highway, and the only thing that stopped me was the driver I would have taken with me. I started hormones nine days later. They did not fix my marriage or my finances. They quieted the noise enough that I could hear myself think for the first time in five decades. People who come that close to not being here, and then make it, do not have much patience left for being told to prove they belong in the water, season after season, by people who will never have to.

I did not enlist for any of this. Like most trans people I know, I would take a quiet life in a heartbeat: coffee strong enough to matter, a band that still books the occasional gig, a stack of books I keep meaning to finish. What I got instead was a standing invitation to argue for my own ordinariness, on cable panels, in legislative hearings, and in the comment threads under articles like this one. The weariness people hear in trans voices is not fragility, and it is not a debating posture. It is the specific fatigue of anyone asked to re-prove a settled thing, on someone else’s schedule, with no end date offered and none intended.

The ponds stayed open. That matters, and it deserves to be said without hedging: a community was asked, it answered clearly, and the water is still there for everyone it was there for last week. The lawyers will be back in the autumn. The swimmers will be back tomorrow. A pond is just a pond, until your right to get into it becomes the thing worth a decade of court dates. Then it is everything. So you swim anyway.

Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

On June 4, OpenAI shipped a feature called Dreaming. It lets ChatGPT update what it remembers about you without being asked, and revise those memories on its own as your life moves. The company that built the fastest-growing app in history now runs a quiet background process whose job is to keep a current file on you.

The headline that week said something else. ChatGPT has surpassed a billion users, making it the fastest app ever to reach that milestone. The number is real enough, but it is not OpenAI’s number. It comes from Sensor Tower, a market intelligence firm, via Reuters, and it counts monthly active app users. OpenAI’s own most recent figure, from February, was 900 million weekly active users. Both can be true at once. The distinction matters. A company that is careful about which metric it claims is a company you should read just as carefully when it tells you what its memory feature does.

So let me tell you what it does, and why the cheering around it should make the back of your neck prickle.

The people building it think memory is the whole point.

Sam Altman has been describing the destination for a long time. At a Sequoia event last spring, he laid it out: he wants a model where every conversation, every book, every email, everything you’ve ever looked at is loaded in, and, in his words, “your life just keeps appending to the context.” He’s noted that younger users already treat ChatGPT like a life advisor and don’t make decisions without it. Dreaming is a step toward that model. OpenAI’s own framing is plainer. Memory, the company says, is central to “knowing you, helping you, and doing more for you.”

The competitors are building the same thing in different ways. Google folded memory into Gemini. Anthropic shipped it to everyone in March and even let people import their history from ChatGPT and Gemini, and described its version, at least at first, as a recall you trigger rather than a profile it assembles on its own. The contrast is the interesting part. Two companies looked at the same feature and made opposite bets about how much of the remembering should happen behind your back. OpenAI bet on automation. Dreaming runs when you’re not looking, and it rewrites itself: the system’s own example is quietly updating “You’re going to Singapore in July” to “You went to Singapore in July 2026” once the trip is over.

Anthropic’s product chief, Mike Krieger, has pitched memory as the road to “sustained thinking partnerships” that evolve over months. That is the case in its best clothes. Not a dossier, a relationship. And on a good day, it feels like one. The model that remembers your project, your deadline, the argument you were building last Tuesday is genuinely more useful than the amnesiac that greeted you fresh every single time. Nobody is wrong about the convenience. The fight is over the price.

The people who study this for a living are less charmed. A 2026 study presented at the field’s main human-computer interaction conference looked at how people actually live with ChatGPT’s memory and found a paradox. The feature most users value is the one they can least see into or control. You like that it knows you. You cannot view the whole of what it knows, and you cannot be sure you’ve deleted any of it. OpenAI says its newest memory recalls facts about you far more accurately than last year’s version did, and I believe the trend. But those scores are the company grading its own homework. No independent audit of the new system existed the week it launched. You are being asked to trust a black box about how faithfully it keeps a copy of you.

That last part is not a feeling. It is a technical fact, and it is where the cheering stops.

How it haunts us

This is the part that should change how you use the thing.

The feature is not evil. The architecture around it is built to keep, and keeping is the opposite of forgetting; you have quietly relied on it your whole life.

When you delete a ChatGPT conversation, you have not necessarily deleted it. We know for a fact that a federal court made OpenAI prove it. In the consolidated copyright case brought by The New York Times and other publishers, a magistrate judge ordered OpenAI in May 2025 to preserve and segregate output logs that would otherwise be erased, including chats users had asked to delete. The order was later narrowed. The court still required OpenAI to hand over 20 million de-identified user conversations to the other side, and in January, a district judge affirmed the order, reasoning in part that users had “voluntarily submitted” their messages when they typed them.

Read the mechanics of that twice. Your delete button cleared your screen. It did not clear the record. A court reached past your deletion, past OpenAI’s own retention policy, and pulled twenty million conversations into a lawsuit, on the logic that you gave them away the moment you hit enter.

Now layer a memory system on top, one whose job is to keep a living profile of you, and the problem compounds. Deleting a memory in the settings page is not the same as the model forgetting. OpenAI’s own help pages admit it: saved memories sit separately from your chats, so deleting a chat does not delete what was learned from it, and scrubbing one detail can mean chasing it through every place it appears. Go down to the training layer, and forgetting becomes nearly impossible. Researchers call the field “machine unlearning,” and the honest summary is that nobody has solved the problem of proving a model has truly forgotten a specific person. A 2018 paper put the whole conflict in its title: humans forget, machines remember. Its authors concluded it may be impossible to honor the legal right to be forgotten inside an AI system at all.

That right is not hypothetical. Article 17 of Europe’s data law gives you the right to have your personal data erased. A statute can command erasure. It cannot make a neural network obey the laws of physics.

And do not count on regulators to close the gap. Italy’s privacy authority fined OpenAI fifteen million euros in 2024 over how it handled personal data. This spring, a Rome court threw out the fine, not on the merits but on a jurisdictional technicality, and pointedly declined to rule on whether the privacy violations had actually occurred. The substance is still undecided. The enforcement evaporated anyway.

Put the pieces together, and a grim sentence assembles itself. The data outlives the conversation; the model cannot cleanly forget; the regulator’s fine was vacated on a technicality; and a court has already shown it can pull the logs. One commentator summed up the era in a headline: the right to be forgotten is dead, and data lives forever in AI. That is too tidy to be the whole truth. It is not too tidy to be a warning.

Memory is a target. Security researchers have already shown that a poisoned web page or document can write hidden instructions into ChatGPT’s long-term memory, causing them to persist across every future conversation and quietly forward what you say to a stranger. OpenAI patched that specific exploit. The category does not patch. A permanent store of your private context is a permanent prize for anyone who wants it, whether a hacker, a divorce lawyer, an immigration officer, or a future government that decides your old questions are evidence.

There is a quieter use than the breach, and it asks no one to break in. The company already holds the record, and it has begun testing what to do with it. OpenAI has trialed ads inside ChatGPT, personalized based on a person’s own chats and memories. Hold that next to Altman’s line about your life appending to the context, and the business model snaps into focus. A system that knows what you fear and what you are ashamed of is a system that can sell to you, or sell you, with a precision no billboard ever managed. Profiling was always the destination. Memory is the part where it gets good at it.

The shape of it is familiar.

Step back, and you have seen this before. Shoshana Zuboff named it surveillance capitalism: a business that treats your lived experience as free raw material, converts it into data, and builds an asymmetry of knowledge and power from that data, which she called unprecedented in the human story. Persistent AI memory is that logic finishing its sentence. The inference machine used to guess at a demographic. Now it remembers your name across the years and revises the file as you sleep.

I have spent a lot of words on who gets to hold a person’s record. When my state moved to expand its surveillance of trans residents, I wrote in “The Anatomy of Erasure” and “Seal the Records” about why a government’s permanent file on a class of people is a weapon and not a filing cabinet. The corporate version is not gentler. It is worse in one specific way. The state needs a subpoena to access a company’s records about you, and, as the OpenAI case just demonstrated, the subpoena works.

I keep finding the same pattern when I write about this. A record gets built for one stated reason, and then the reason quietly widens. The list meant to protect becomes the list used to find. It takes no villain. It takes a database, a new administration, and a definition of threat that grew as no one reread the fine print. An auto-updating memory is that fine print, rewriting itself overnight, on a server you do not own.

What we actually want it to remember

So what do we want it to remember?

Not nothing. The appeal is real. A tool that remembers that you take your notes in metric, that your kid’s name is Sam, and that you don’t have to reintroduce him every single morning saves the small daily tax of having to start from scratch. That is worth having.

What we want is the memory we chose. We want it to hold the preferences we hand it on purpose and to let the rest go. We want an expiration date. We want to carry it with us when we leave, the way Anthropic’s import tool gestures at, and we want leaving to be real, deletion that reaches the model and not just the screen. We want a mode where the thing simply does not remember, and we want to trust that the mode is telling the truth.

None of this is utopian. Time decay of old memories, an export file you actually own, a no-memory mode that is genuinely empty, deletion verified down to the weights: these are engineering problems, not laws of nature. The companies have not been pushed to prioritize them. The version that remembers everything wins the demo.

Altman, to his credit, has floated one piece of this. He has argued for what he calls AI privilege, a legal shield around your conversations with a model, similar to the one around what you tell a lawyer or a doctor. It does not exist yet. It is worth building. But privilege protects the conversation. It does not answer the harder question, which is whether a machine should be assembling a version of you that you never sat down to write, that edits itself overnight, and that outlives your decision to be done with it.

The teenager typing the questions she cannot ask anyone else is not building a profile. She thinks she is having a conversation. The distance between those two things is the entire problem. We get to decide, with the defaults still soft and the laws still being drafted, whether the most intimate record ever kept of ordinary people belongs to the people inside it.

We should decide on purpose. The machine is already deciding for us, every night, in its sleep.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

The news broke on a Monday, and by Tuesday, half the timeline had promised to read the S-1. There is no S-1 to read. Not yet, and possibly not for months.

Here is what actually happened. On June 1, 2026, Anthropic announced that it had confidentially submitted a draft registration statement on Form S-1 to the Securities and Exchange Commission. The word doing the work in that sentence is “confidentially.” The company told the public that it filed. It did not show the public what it filed.

That distinction is the whole story, so let’s take it slowly.

A draft nobody outside the SEC can see

When a private company wants to sell stock to the public, it files a registration statement called an S-1. The S-1 is the document everyone wants. It’s where a company has to write down the things it would rather not say out loud: how much money it makes, how much it loses, who owns what, and everything that could go wrong.

A 2012 law called the JOBS Act added a side door. Companies under a certain size can hand the SEC a draft S-1 in private, let the agency’s staff review it and send back comments, and fix the embarrassing parts before anyone else reads a word. The public version comes later. By rule, it must be publicly on file at least 15 days before the company starts pitching investors in person.

So the confidential filing is real. It’s a genuine step, not a press release wearing a costume. What it is not is readable. Anthropic put out a one-paragraph notice under an SEC rule that exists precisely to let a company confirm it filed without saying anything else. You will get the actual numbers when the redacted draft goes public. Until then, anyone walking you through “Anthropic’s prospectus” is walking you through a document they have not seen.

The number everyone repeated.

What we do know comes from a different event three days earlier. On May 28, Anthropic said it had closed a $65 billion Series H funding round, valuing the company at $965 billion.

Read that twice. Sixty-five billion dollars raised in a single round. A $965 billion valuation, which is the kind of number that stops meaning anything once it gets that big, so here are two handholds.

Goldman Sachs, the investment bank, is worth around $300 billion in public markets as of early June. Anthropic’s private valuation is roughly three times that. The firm that has been a synonym for Wall Street money for a century and a half would fit inside this AI company three times over, on paper.

Second handhold. OpenAI, Anthropic’s larger rival, was valued at $852 billion in its own round earlier this year. Anthropic just leapfrogged it. Only about a dozen public companies on earth are worth a trillion dollars or more, and a $965 billion company would sit just under that ceiling.

Now, the part the headlines skip. A private valuation and a public market value are not the same animal in a different coat.

Why $965 billion does not convert cleanly

A private valuation is the amount a handful of investors agreed to pay for a specific slice of the company, usually preferred shares that come with protections that ordinary shareholders never get. If the company stumbles, those investors often get paid back first. The headline figure is the price of the last shares sold, multiplied by the number of shares, as if they were all worth the same. They are not all worth the same.

A public market value is the opposite. It’s whatever millions of strangers will pay for the stock that minute, with no protections and no floor, repriced every time the market is open. The $965 billion is a private agreement. The IPO will be a public referendum. Sometimes the referendum agrees with the agreement. Often it does not.

What does it mean if you use the products?

For most people reading this, the practical question is simple. Does Anthropic going public change the product you use, or the price you pay for it?

Honest answer: not on day one, and the long-run direction cuts both ways.

A public company answers to public shareholders every quarter. That pressure can push a company toward the boring, profitable things and away from the expensive research bets. Research is a cost now and a maybe in ten years. It can run the other way, too. A war chest from a successful IPO buys more compute, more people, and time.

What changes for certain is visibility. Right now, Anthropic tells you what it wants you to hear. A public company has to file audited financials four times a year. You will learn, in numbers it cannot spin, whether the products you depend on are sold at a profit or burned at a loss to win the market. That tells you more about a tool’s future than any keynote does.

The one figure already public is a tell. Anthropic’s chief financial officer said its run-rate revenue crossed $47 billion. Revenue is not profit. The S-1, when it surfaces, will show the gap.

The section to read first

When the redacted S-1 does go public, skip the founder’s letter and the mission language. Go to two sections.

“Risk Factors” is where the company is legally required to list everything that could sink it, written by lawyers whose job is to make sure nobody can later claim they were not warned. “Legal Proceedings” is where active lawsuits live.

For an AI company, that is where the real reading is. These models are trained on enormous amounts of text and images, and the question of who owns that material and whether using it is permitted has produced a wave of litigation across the industry. How a company describes that exposure and what suits it names tell you more about the real risk than the valuation ever could. Read those pages against what the company has said in public, and watch for the gaps between the two.

That is the document worth waiting for. Not the valuation. The fine print.

Where this actually stands

A few things are confirmed, and a few are not, and the difference matters.

Confirmed: the confidential draft S-1, filed June 1. The $65 billion round and the $965 billion valuation closed on May 28. Confirmed by Anthropic’s own announcements, not by a tip.

Not confirmed: the timing and the venue. You will see “October” and “NASDAQ” reported with confidence. Those come from analysts and from reporting about internal discussions, not from Anthropic, which said in its own notice that timing depends on market conditions and that the price and share count are not set. An IPO target is a plan. Plans on this scale move.

So treat the date as a rumor and the filing as a fact. Mark your calendar in pencil. And when the real document lands, read the risk factors first.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

Sources

• Anthropic, “Anthropic confidentially submits draft registration statement for proposed IPO,” company newsroom, June 1, 2026.

• Anthropic, “Anthropic raises $65B in Series H funding at $965B post-money valuation,” company newsroom, May 28, 2026.

• Jordan Novet, “Anthropic tops OpenAI as most valuable AI startup, nears $1 trillion valuation in latest round,” CNBC, May 28, 2026.

• Manya Saini, “AI giant Anthropic confidentially files for US IPO as investors bet big on AI future,” Reuters, June 1, 2026.

• “Anthropic Pulls Ahead of OpenAI in High-Stakes Race for IPO Riches,” Bloomberg, June 1, 2026.

• “Anthropic files to go public,” TechCrunch, June 1, 2026.

• “Anthropic Makes Confidential IPO Filing,” The Information, June 1, 2026.

On June 4, 2026, prosecutors in Budapest dropped the criminal case against Mayor Gergely Karácsony for organizing a Pride march the government had banned. The charge fell apart for a plain reason: the law behind the ban had already been struck down. What the wire stories mostly skipped is the longer story underneath that wins, which is not about a court or a mayor at all. It is about a coalition that Viktor Orbán spent sixteen years assembling by mistake.

Gender as “symbolic glue”

Orbán’s strategists had a word for the thing they thought they were fighting. They called gender “symbolic glue, ”a phrase scholars of the Hungarian and Polish right used to describe how the movement bound its scattered targets into one. Feminism, trans people, sex education, same-sex marriage, abortion access, gender studies: file them all under “gender ideology,” and they become a single foreign menace a government can campaign against. The glue was meant to hold the enemies together so they could be fought as a single unit. It worked. That was the miscalculation.

One door, marked “the family”

Watch how the measures landed. In 2018, the government revoked accreditation for gender-studies master’s programs at ELTE and Central European University, with Deputy Prime Minister Zsolt Semjén dismissing the field as an ideology rather than a science. In May 2020, Parliament refused to ratify the Istanbul Convention on violence against women, casting the treaty as a vehicle for gender ideology and migration, and ended legal gender recognition for trans people the same month. In 2021, came the “child protection” law that lumped LGBTQ content in with pedophilia. Every one of those blows reached feminists and queer people through the same door, marked “the family” and “protecting children.” Aim a single weapon at two groups, and you teach them, fast, that they share an enemy.

Older than Orbán

The alliance that answered was not a hashtag. Hungarian feminist and queer organizing each long predate Orbán. NANE, the women’s anti-violence group that runs the country’s domestic-violence helpline, has worked since 1994; the legal-aid group PATENT since 2006. Háttér Society, the largest LGBTQI organization in the country, was founded in 1995, and the Labrisz Lesbian Association in 1999. When the 2021 law passed, the Hungarian Women’s Lobby, NANE, and PATENT put out a joint statement condemning it, insisting that “education is not the same as promotion.” Háttér and Labrisz had co-founded Hungarian LGBT History Month back in 2013. The collaboration isn’t a memory: into 2026, Háttér has been running a project it calls “Diverse Feminism,” built with the Nőkért Association, to intentionally keep the feminist and queer movements in the same room.#

Banned, and held anyway

Then the government banned Pride outright. A March 2025 amendment to the assembly law and the fifteenth amendment to the Constitution, passed that spring, redefined a person as male or female, elevated “child protection” above the right to assemble, and authorized facial-recognition surveillance of anyone who showed up. The civil society response was coordinated and fast. Amnesty International Hungary, Háttér, the Hungarian Helsinki Committee, and the civil-liberties group TASZ challenged the police bans. They published legal guidance for marchers, while Karácsony reframed the event as a municipal gathering, which he argued the police had no power to prohibit. On June 28, 2025, roughly 100,000 people marched anyway, with organizers putting the number far higher. It was the largest Pride in Hungarian history and one of the largest demonstrations against the government in years. A solidarity statement signed by 71 organizations put women’s rights groups such as the Hungarian Women’s Lobby, EMMA, and MENŐK on the same page as queer and civil liberties NGOs.

“Neither freedom nor love can be banned in Budapest.” — Mayor Gergely Karácsony, at the June 28, 2025, march

Vindication in two acts

The vindication arrived in two acts. On April 21, 2026, the Court of Justice of the European Union ruled in Case C-769/22 that the 2021 law breached EU law, and for the first time in its history, found a standalone violation of Article 2 of the Treaty on European Union, the clause naming the Union’s founding values. That is the EU’s own court in Luxembourg, not the human-rights court in Strasbourg that gets confused with it. Six weeks after that ruling, the charges against Karácsony and against Géza Buzás-Hábel, the gay Roma teacher who organizes Hungary’s only rural Pride in Pécs, were dropped on the same day. Police had already cleared the 2026 march to proceed on June 27.

Weather, not a change of heart

Here is the turn, and it matters. The coalition did not win an ally. It outlasted an enemy. Péter Magyar’s Tisza party won the April 12 election, and he was sworn in as prime minister on May 9, 2026, ending Orbán’s sixteen years in power. Magyar speaks about freedom and assembly in the abstract. He has not endorsed Pride, and he has not moved to repeal a single one of the laws Orbán built. The statutes that stripped trans recognition and banned queer content from minors are still on the books. Buzás-Hábel said it without varnish: the main opposition remains quiet on LGBTQ rights, so the movement will have to keep up the pressure. A change of government is weather. It is not a change of heart.

I write this from Sioux Falls, South Dakota, where my own statehouse files its own bills naming people like me as a problem to be managed. I came out at fifty-three, which means I have watched a lot of organizing in my own country splinter along exactly the seam Orbán went looking for, the one where someone whispers that the trans people are costing the women’s movement its credibility, or that the feminists were never really allies. The Hungarian record is the rebuttal. The lesson is not that a court will ride in and save you, since courts arrive late, when they arrive at all. The lesson is the glue. Orbán’s people mixed it to bind their enemies into one convenient target. It dried, and it held, and it bound those enemies to one another instead. The cameras are still bolted to the lampposts in Budapest. Pride is still on the calendar for June 27. They wanted one enemy. They built one movement.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

On May 19, 2026, Google walked onto the I/O stage and said its new Gemini 3.5 Flash beats Gemini 3.1 Pro on almost every benchmark that matters. Five days later, three independent labs re-ran those benchmarks. The numbers matched to the decimal. And the model still came out looking worse than the keynote suggested.

Both things are true at once. That gap, between an accurate number and an honest one, is the whole story.

What actually shipped

First, a correction worth making out loud. It is the exact kind of error this piece is about. Gemini 3.5 Flash did not launch in early June. It shipped on May 19 at Google I/O 2026 and was generally available from day one, with no preview window. If you read that it launched this month, you read an aggregator that copied another aggregator. The original record is one tap away, and it disagrees.

Flash is the first model in Google’s new 3.5 family. The bigger sibling, Gemini 3.5 Pro, was still in internal testing on launch day, with a public release promised for the following month. So, the model everyone is arguing about is the small, fast one, pushed out ahead of the flagship.

What “agentic” even means

Google’s pitch for Flash isn’t that it writes a nicer paragraph. It’s that the model can do a job. Google calls it its “strongest agentic and coding model yet.” An agentic workflow is the difference between an assistant that answers a question and one that carries out a task: it plans the steps, writes and runs code, checks its own output, hits an error, tries again, and keeps going, sometimes for hours, with little supervision. Google’s demo had two of these agents cooperating to build a working game in an afternoon.

That’s why Google leads with speed. When a model only answers you, a half-second delay is invisible. When a model runs an agent loop that calls itself four or five times for every request, the delays stack, speed stops being a nicety, and becomes the bill. So Google’s headline claim is that Flash produces text four times faster than other frontier models, though it never names which ones.

There’s a measurement under that word “faster.” Models read and write in tokens, chunks of text roughly the size of a short word or a piece of one. Output tokens per second is just how fast the model types. Flash runs at around 280 of them per second, according to independent measurements, which is quick for its intelligence class. On that count, Google isn’t exaggerating.

The numbers were real.

Here’s the part that separates this from a fraud story. Google DeepMind handed pre-release access to Artificial Analysis, an independent evaluation outfit, before launch. Artificial Analysis ran its own suite and confirmed Google’s headline scores. Terminal-Bench, MCP Atlas, the multimodal tests: the figures held. On MCP Atlas, a measure of how well a model handles tool calls, Flash genuinely leads the field. Nobody caught Google inventing a number.

That matters. The easy version of this article would be “Google lied,” and it would be wrong.

What got left off the slide

What Google did was quieter and far more common. It chose the comparison and the metrics.

The comparison first. Google benchmarked its shiny new Flash against its own previous-generation Gemini 3.1 Pro. Not against the models it competes with in June 2026, which are OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7. Beating last year’s house model is a real achievement. It isn’t the same achievement as beating the current field, and the keynote let the room hear the second thing, having proved only the first. One independent roundup put it cleanly: the agentic story is real, and the frontier-intelligence framing is conditional.

Then the metrics. The same independent labs published three numbers Google didn’t put on a slide.

Flash lands somewhere around eighth place on Artificial Analysis’s overall intelligence index, behind both GPT-5.5 and Claude Opus 4.7. Strong. Not the top of the chart. The exact rank wobbles between fifth and eighth depending on the day and the configuration, which is itself a lesson in how soft these rankings are.

Its hallucination rate, how often it states something false with confidence instead of admitting it doesn’t know, is 61 percent. That’s a large improvement over the previous Flash, which sat above 90. It’s still a coin flip you wouldn’t want to build a knowledge product on without guardrails.

And the cost. This is the one that stung the developers. Flash is cheaper per token than the old Pro. It’s more expensive per job. Google tripled the token price compared with the previous Flash, and the new model is chatty, so it burns more tokens thinking out loud. Running Artificial Analysis’s full test suite on Flash costs about $1,552, roughly 75 percent more than running the very Gemini 3.1 Pro it claims to beat. A lower sticker price on a thirstier engine isn’t a lower fuel bill.

The independent developer Simon Willison flagged the same price jump on launch day and framed it as part of a wider pattern. All three major labs have started probing how much their API customers will tolerate paying. Flash is one data point in a trend, not a Google quirk.

Why does this keep happening

None of this is unique to Google, and Google’s version is mild. The instructive comparison is the one that actually was a scandal.

In April 2025, Meta launched Llama 4 and posted a glittering leaderboard score. The number came from a special chat-tuned variant submitted to the rankings, not the model the public could download. When testers ran the real weights, the model fell dozens of places. For months, Meta waved this off. Then, in a Financial Times interview reported at the start of January 2026, Meta’s departing chief AI scientist, Yann LeCun, said the results had been “fudged a little bit” and that the team used different versions of the model across different benchmarks to achieve better scores.

That’s the far end of the spectrum: a different model on the scoreboard than in your hands. Google didn’t do that. Google ran the real model and reported real numbers. It picked the flattering opponent and skipped the unflattering stats. The two sit on the same spectrum, which is the point. The whole industry has an incentive to choose its own framing. Stock prices and developer mindshare align with launch-day benchmarks. The measure became the target, and a measure that doubles as a target stops measuring much.

What to do with all this

If you’re picking a model, the lesson is small and durable. Read the comparison, not the number alone. Ask who the new model was tested against, and whether that’s who you’d actually be choosing between. Ask which benchmarks were shown, and guess at which ones weren’t. And if cost matters, price your own job, not the per-token rate. A fast, cheap-looking model that talks too much can quietly outspend the expensive one.

If Flash is the model you’re building on, it earns its place on agentic, tool-heavy, multimodal work, where it genuinely leads. For the hardest reasoning and for long-document recall, the current frontier models still beat it, and Gemini 3.5 Pro isn’t out yet. Test it on your own workload before you believe anyone’s slide, including this one.

The honest headline was available to Google the whole time. Fastest model in its class, real agentic gains, priced higher than it looks, behind the frontier on raw reasoning. That version would have survived contact with the independent labs intact. Google reached for a better story than the truth, and the truth showed up five days later anyway. It always does.

Author Note. Grace Ann Hansen is an independent researcher and writer, and an MBA & PhD graduate student in health informatics and artificial intelligence. She is also a published author, a professional musician, a gymnastics coach, and a queer transgender woman living in Sioux Falls, South Dakota. All interpretation, argument, and prose are her own. Correspondence concerning this article should be addressed to Grace Ann Hansen at grace@graceannhansen.com.

Sources

• Koray Kavukcuoglu, Jeff Dean, Oriol Vinyals, and Noam Shazeer, “Gemini 3.5: frontier intelligence with action,” Google, May 19, 2026.

• “With Gemini 3.5 Flash, Google bets its next AI wave on agents, not chatbots,” TechCrunch, May 19, 2026.

• “Gemini 3.5 Flash: everything you need to know,” Artificial Analysis, May 2026.

• “Gemini 3.5 Flash, 5 Days In: Independent Eval Roundup,” Digital Applied, May 2026.

• Simon Willison, “Gemini 3.5 Flash: more expensive, but Google plans to use it for everything,” May 19, 2026.

• “Meta accused of Llama 4 bait-n-switch to juice LMArena rank,” The Register, April 8, 2025.

• “‘Results Were Fudged’: Departing Meta AI Chief Confirms Llama 4 Benchmark Manipulation,” Slashdot, reporting a Financial Times interview, January 2, 2026.

I spent most of 2023 telling myself ChatGPT was a toy. I had 35 years of IT consulting and project management behind me by then, inside pharmaceutical companies, defense contractors, the Department of Defense, and a few household names in tech, entertainment, and soft drinks, and I had watched plenty of overhyped technologies arrive with a press release and leave with a shrug. A chatbot that wrote limericks looked like a party trick.

Then I read the threat reports, and the party trick was holding a crowbar.

What the bad guys do with it

In December 2024, the FBI’s Internet Crime Complaint Center warned that criminals exploit generative AI “to commit fraud on a larger scale” and to make their schemes more believable (FBI IC3, 2024). Microsoft and OpenAI had already named names. In February 2024, the two companies disclosed five state-affiliated groups, tied to Russia, North Korea, Iran, and China, caught using large language models to scout targets, draft phishing lures, troubleshoot malicious code, and translate the results into cleaner English (Microsoft Threat Intelligence, 2024).

By 2025, the numbers had teeth of their own. Microsoft’s Digital Defense Report measured AI-automated phishing at a 54 percent click-through rate, compared with 12 percent for the human-written kind (Microsoft, 2025). In November 2025, Anthropic disclosed a campaign it assessed as Chinese state-sponsored, one that manipulated Anthropic’s own coding model to run 80 to 90 percent of an espionage operation against roughly 30 targets, with humans stepping in at only a handful of decision points (Anthropic, 2025).

Nobody on the attacking side asked permission. Nobody on the attacking side held a panel discussion about whether the technology was ethical. They picked up the crowbar and went to work.

The data I was paid to guard

When COVID froze the consulting world, I took a job in my hometown as a data engineer and an analytic systems administrator at a health system, and as the technical project manager for anything that touched those platforms. Every table in every database I kept running was somebody’s diagnosis, somebody’s insurance claim, somebody’s kid.

Healthcare is the sector that criminals like best. A 2025 study in JAMA Network Open reported 6 million Americans affected by healthcare data breaches in 2010 and 170 million in 2024, with hacking accounting for 88 percent of records exposed across those years (Jiang, Ross, and Bai, 2025). IBM’s annual breach report priced the average healthcare breach at $7.42 million, the most expensive of any industry it studied (IBM, 2025). I read those reports the way you read weather radar when your house is in the path of a tornado.

So I did what a project manager does when the scope changes. My employer handed a group of us subscriptions to every major AI platform and told us to learn what we did not know: ChatGPT, Gemini, Claude, Grok, and the foreign models DeepSeek, Doubao, Qwen, and Baidu. For a year, I studied artificial intelligence on nights and weekends and earned industry certifications along the way.

Read more

GPT-4 passed the bar exam at the 90th percentile of human test-takers (Katz et al., 2024). GPT-4 has, in well-documented public sessions, confidently insisted that 9.11 is larger than 9.9. The same system does both things. If you go looking for evidence that it understands, you can find it. If you go looking for evidence that it does not, you can find that too.

This is the question Part 1 cannot answer, and this lesson is going to be careful with it. We will close Part 1 by naming the question, drawing the map of the field we have been crossing, and looking honestly at how the smartest people in the field disagree about whether anything inside a chatbot deserves to be called comprehension. The point of the capstone is not to give you the answer. The point is to make sure you do not borrow somebody else’s answer without realizing you have done it.

Read more

Two AI systems are running as you read this sentence. One is at your bank, watching your debit card. If you swipe something it thinks is fraud, it locks the card before you finish dinner. The other is in your browser. If you type a question into ChatGPT, it writes you a paragraph. Both are AI. They are completely different shapes of tool, and the difference matters more than the news coverage suggests.

This is the last triage move you need before Part 1 closes. Within the umbrella of “AI,” the largest split is between systems that predict (the older, quieter, much bigger industry) and systems that generate (the newer, louder, much-discussed slice). The same recipe of data, compute, and algorithms from last week powers both. The output shape is different. The risks are different. The way to read headlines about each is different.

Read more