The Salem Witch Trials, But Make It AI

The Setup

A sophomore at Palo Alto High School — a kid with competitive college ambitions — writes an essay about The Crucible, Arthur Miller's play about the Salem witch trials. You know, the one about mass hysteria, false accusations, and an innocent person destroyed by a system that couldn't tell fact from fever dream.

He submits it October 30, 2025. Two weeks later, Turnitin runs it through whatever statistical alchemy these systems use and comes back with a verdict: 76% AI-generated.

The teacher, Sarah Bartlett, invokes what PAUSD calls a "non-punitive" cheating policy. Translation: you didn't technically cheat, but we're going to treat you like you did anyway. The kid retakes the assignment in class. He gets a D. His overall grade drops to a C.

This is the part where you'd expect the kid to roll over. He didn't. His dad — Takashi Kato, who works in AI, which is important — picked up the phone and called a lawyer instead of a therapist.

The Irony

The Crucible is about a town that destroys an innocent kid because the adults in charge can't tell the difference between a frightened teenager and a witch. Now a kid who wrote about The Crucible is being destroyed by a system that can't tell the difference between a frightened teenager's essay and an AI output.

Kato submitted 1,162 pages of Google Docs revision history — drafts, edits, notes, the messy iterative evidence that a human actually wrote the thing. Turnitin looked at all of that and said: nah, still 76% robot.

But here's the number that should make you furious: Turnitin itself acknowledges a ±15 point variance on its own scores. So 76% AI could be 61%. Or it could be 91%. They don't know. And in case that's not damning enough, Turnitin's own published guidelines say the tool "should not be used as the sole basis for adverse actions against a student." PAUSD appears to have done exactly that.

And then there's the part that sounds like dark comedy: After the family challenged the result, Assistant Principal Jerry Berkson had the student's handwritten in-class rewrite typed up by a secretary and run through Turnitin again — without the family's consent. Using the same broken detector to validate itself. That's not a cheating policy. That's a circular firing squad.

The kid's essay was about people being wrongly accused of things they didn't do.

This is either the universe having a laugh, or Arthur Miller posthumously winning the argument.

The Valley Weighs In

Elon: So a kid wrote about the Salem witch trials and the school responded by... conducting one.

Sam: I mean, the false positive rate on these detectors is actually —

Elon: Is none of your business, Sam. You built the thing that writes the essays AND the thing that flags them. You're selling the disease and the cure.

Jensen: He's not wrong. My GPUs run both sides. I literally cannot lose.

Sam: That's... not something to brag about, Jensen.

Jensen: I'm not bragging. I'm diversifying.

Zuck: Can we talk about the ±15 variance? Seventy-six percent AI could be sixty-one. It could be ninety-one. The margin of error IS the entire range of outcomes.

Bezos: That's a worse error rate than our customer reviews. And we let people rate cat litter.

Bill: The real issue is the district has no AI policy. No training, no standard. They handed teachers a black box and said "good luck."

Elon: And then — this is my favorite part — they had the kid handwrite the essay again, typed it up on a secretary's computer, and ran it through Turnitin. Again.

Sam: They didn't.

Elon: They did. Used the broken detector to validate the broken detector. That's not due process. That's a circular firing squad.

Zuck: Wait. His dad works in AI?

Jensen: Works in AI and filed a hundred-and-fifty-million-dollar lawsuit. That's not a parent-teacher conference. That's a plaintiff.

Sam: And the kid's grade is still a C?

Bezos: Still a C.

Bill: I wrote four thousand words about this and somehow said less than Elon just did in one sentence.

Elon: That's because I deleted four thousand words and kept the one that mattered.

The Real Issue

Here's the part PAUSD doesn't want to talk about in their June 12 response — denying all claims and citing teacher grading authority under California Education Code § 49066, which gives teachers "exclusive, autonomous grading authority" absent clerical mistake, fraud, bad faith, or incompetency:

There is no district-wide AI policy.

Teachers are making it up. Some use Turnitin. Some use GPTZero. Some use a vibes-based assessment that amounts to "this doesn't sound like a 15-year-old." There's no standard. There's no training. There's no appeal process worth mentioning. Just a software flag, a teacher's gut check, and a kid's grade hanging in the balance.

Kato's lawsuit is, at its core, a challenge to that chaos. When you have a tool that admits a ±15 point variance, flags 61% of non-native English speakers' essays as AI (per Stanford HAI), and disproportionately targets male students by 4–5× — and you give it to teachers who have no framework for interpreting it — you don't have a cheating policy. You have a liability.

The case is Kato v. PAUSD, 5:26-cv-04078 in the Northern District of California. Filed May 5, 2026. Fourteen claims including Title VI (national origin discrimination — Kato is Asian and multilingual), Title IX (sex discrimination — the detector's bias against male students), retaliation, and due process violations. Case management conference set for October 8, 2026. At least one other Asian student in the same class received the same treatment — Turnitin flag, forced rewrite, D grade, semester grade dropped. No amicus briefs yet.

Someone should probably file one.

$150 million is a lot of money (it's the prayer for relief figure, not a damages calculation — but it's a big number to put in a headline and an even bigger target on the district's legal budget). But the cost of not having an AI policy isn't just lawsuits — it's every kid who gets a D because they wrote in a style an algorithm didn't like.

The $150 Million Question

Prayer for relief in Kato v. PAUSD — the cost of not having an AI policy

If PAUSD loses — if a school district can be held liable for acting on a false positive from an AI detection tool — the entire ed-tech industry has a very expensive problem. Turnitin, OpenAI, Google, every company that told schools "trust our AI" is watching this case.

The kid wrote about injustice and false accusations. He became a case study in both.

His grade is still a C.

That's a Crucible worth writing about.

— The Panel