A Single Death Is a Tragedy. Fifty Is a Threshold.

Illinois did something real this month, so let me say the nice part first before I rain on their parade. SB 315 passed the House 110 to 0. Then the governor signed it. It's the third state to put guardrails on the most powerful AI models on earth and the first to demand an actual independent audit. Historic. Genuinely. I will take it.

Now let me ruin it.

Open the bill and search for the definition of "catastrophic risk," because that one phrase decides who this law was built to protect, and the answer is sitting right there in templatized font used for legislative print. Here's what officially counts as a catastrophe. The death of, or serious injury to, more than 50 people. Or a billion dollars in property. Or a chemical weapon, a runaway cyberattack, or a model that slips its leash.

More than 50 people.

Marinate on that number, because someone chose it. Someone in a room, with coffee and a redline, decided that 50 was the line. 49 dead from a frontier AI system in a single incident? Not catastrophic. Technically. Legally. 49 moms, dads, daughters, brothers, the person somebody calls when the car breaks down, and the statute shrugs. Sorry, you didn't clear the threshold to matter. Try again with a bigger pile.

There's a line that history attributes to Stalin, fairly or not. A single death is a tragedy. A million deaths is a statistic. We used to quote that to remind ourselves what monsters sound like. Illinois just wrote a gentler version into law and called it a win. A single death is a tragedy. 51 is a compliance trigger. But 50? That's a rounding error.

You can't reduce people to a body count and then set the count high enough that the bodies stay convenient. Every single one of those 50 is a tragedy that doesn't become more tragic at number 51. The kitchen chair where they used to sit is empty either way. The math doesn't care, which is sort of the whole point of why I'm compelled to write this.

And here's the part that should really get you. The catastrophe the law is bracing for hasn't happened yet but the catastrophe it ignores already did.

Tell the threshold to Michigan, which ran an automated system that falsely accused tens of thousands of residents of unemployment fraud, garnished their wages, seized their tax refunds, and shoved families toward bankruptcy, on the say-so of an algorithm that was wrong the vast majority of the time. Tell it to the disabled Arkansans whose home-care hours got hacked in half by a formula not one person in the building could fully explain. Tell it to everyone in Pennsylvania denied the benefits they had already earned.

Not hypotheticals. Not the future. Not a runaway superintelligence in a think-tank slide deck. Real people, real names, real ruin, already on the books. And under this historic law's definition of catastrophe, every last one of them is a rounding error too. Nobody died in a tidy single incident. No billion-dollar crater. So, based on Illinois law, it doesn't count.

And here's the tell, and once you see it you can't unsee it. Look at who Illinois put in charge. Not a civil rights office. Not a consumer protection agency. The Emergency Management Agency and the Office of Homeland Security. The flood-and-terrorism folks.

That's a filing decision. Illinois opened its drawer of dangers, found artificial intelligence, and filed it under, welp, disaster. Under the thing that arrives all at once with a body count and press conference filled with thoughts and prayers. Not under the thing that has spent a decade quietly grinding through hiring portals and credit models and benefits offices, sorting human beings into yes and no with no appeal, no apology, and no incident report. Because apparently a decade of documented algorithmic discrimination wasn't educational enough.

This is the split nobody in the press wants to name, so I will. There are two stories about AI harm.

• Story one: the runaway model, the engineered pandemic, the machine that decides it doesn't need us.
• Story two: the machine that works exactly as designed and decides you don't belong.

Story one gets the keynotes, the funding, and now the statute. Story two gets the victims. Same technology. Only one of them got a law.

I'm not saying story one is fake, not at all. The people losing sleep over catastrophic risk aren’t fools, and I would rather we worry about the engineered pandemic than not. But notice the trade Illinois just made. It wrote a law for the catastrophe that might happen and went stone cold silent on the catastrophe that is already happening. It protected the future and abandoned the present. And the present is where the people who look like me actually live.

I know this one personally. I asked ChatGPT to turn my photo into an action figure. A Cloud Jedi, lightsaber and all, same prompt half of LinkedIn was running at the time. My white LinkedIn colleagues got crisp little replicas of themselves. I got Jensen Huang's face. The model read the words "tech CEO," looked at me, decided those two things couldn't possibly coexist, and corrected the error. It handed my body back in the correct red dress, but wearing Jensen Huang's face.

Nobody died. No building fell. But, by this law's arithmetic, nothing catastrophic occurred. The machine completely erased and replaced me with who it thought belonged in the chair I built with my own hands. And if it does that to a founder with 3+ decades of technology expertise, who builds and audits these systems for a living, go ahead and ask what it's doing to the kid whose resume it quietly never forwards, the patient whose pain it scores as exaggeration, the family it prices out of a mortgage. When algorithms can't see us, they can't serve us - fairly. And that harm has a name. It’s discrimination. Discrimination at the speed and scale of AI isn’t a smaller problem than a rogue model. It's just a quieter one, and quiet is exactly how it keeps getting left out of these bills.

The people who saw this coming drew the map years ago.

• Joy Buolamwini showed us the facial systems that could not see dark skin.
• Safiya Noble showed us the search engines teaching a generation what Black girls were supposedly worth.
• Cathy O'Neil named the entire machine, weapons of math destruction, models that punish the poor and reward the powerful while hiding behind the word "objective."

None of them were worried about the meteor. They were worried about the spreadsheet. And they were right.

Now here comes the part that turns an oversight into a choice. When the harm shows up in hiring, Illinois already acted. The state amended its own Human Rights Act to ban employers from using AI that discriminates on the basis of a protected class, and that ban has been live, enforceable law since January 1, 2026, 6 months before SB 315 ever reached the floor. A statute already on the books, already working, sitting in the same building that wrote the catastrophe bill.

And we know exactly what it was built to catch, because the federal government already caught a version of it. A tutoring company called iTutorGroup programmed its hiring software to automatically reject women aged 55 and older and men aged 60 and older. More than 200 qualified people, screened out for the crime of having a birthday. One applicant cracked it by sending the same resume twice, once with her real birth date and once with a younger one. Only the younger one got an interview. The EEOC sued. The company paid. That’s algorithmic discrimination with a face, a docket number, and a settlement check, and it is precisely the harm Illinois decided was worth a law.

So, the vocabulary exists. The muscle exists. The state flexed it. Which means when SB 315 defines catastrophe as a body count and strolls right past algorithmic discrimination, that's not a gap somebody forgot to fill. That was a decision about whose harm gets a statute and whose harm gets to keep on waiting.

Here's the twist that should keep folks up at night. The law Illinois wrote for present harm has sharper teeth than the one it just wrote for catastrophe. AI discrimination in hiring runs through the Human Rights Act's own enforcement, which means a person who got hurt has a real path to a remedy. But with SB 315? It hands enforcement to the Attorney General alone, caps the fines, and gives the rest of us no right to sue at all. Yep, you're gonna want to read that again. The catastrophe bill, the historic one, the nation-leading one, is the one with the weaker teeth for the actual victims.

And if you step one foot outside to a company office, to the Medicaid line, the benefits window, the loan desk, and there's no instrument at all. The hiring statute stops at the office door. Michigan could happen in Illinois tomorrow. Arkansas could happen here next week. Not one Illinois law would so much as blink. But, SB 315, the brand-new AI law to take effect in 2027 was written by people who clearly understand this technology, looked at every one of those families and chose to aim the entire apparatus at the future instead.

So, yes, go ahead and celebrate SB 315 for what it is. A real audit. Real whistleblower protection. A real gut punch for the era of "just trust Big Tech." But zoom out, because Illinois didn't pick this lane by itself. Look at Washington.

In June, an Executive Order was signed to build a federal framework for frontier AI. Secure deployment. Voluntary early sneak peeks at new models before the public gets them. A classified benchmarking process so the government can grade the most powerful systems on earth from behind a curtain. The same catastrophe framing again, all the way up, now with a national-security bow on top.

And six months before that (last December), a different Executive Order, aimed straight at the states. It put state algorithmic-discrimination laws on notice, named Colorado's by name, and stood up a Justice Department task force to go hunting for state AI rules to drag into court. The official objection to anti-discrimination law? That it might force AI to produce false results to avoid differential treatment. Sit with that sentence until it stops making sense, because it never will. The federal position appears to be that requiring an algorithm not to discriminate is the same as requiring it to lie.

And they didn't stop at naming it. They made an example of it.

Colorado had passed the first comprehensive AI law in the country, the one built to stop exactly the discrimination I've been describing, with duties of care, impact assessments, and real accountability for decisions about jobs, housing, healthcare, and credit. Industry spent two years calling it too burdensome, which mostly meant most companies never bothered to build what it asked for. Then in April, Elon Musk's xAI sued. The Justice Department jumped in on xAI's side, the first time in history the federal government has gone to court to kill a state AI law. The court froze enforcement. And on May 14, Colorado folded. It repealed its own landmark statute and traded it for a thin notice-and-disclosure shell.

• The duty of care? Gone.
• The risk programs? Gone.
• The impact assessments? Gone.
• The discrimination protections, the entire reason the thing existed? Gone.

And the watered-down replacement doesn't even switch on until 2027. Which is to say, whenever it's convenient.

Here's the part nobody who cheered that rollback wants to say out loud. Gutting the statute didn't gut the harm. The risk didn't get legislated away. It just moved. When an algorithm wrongly denies someone a job, a loan, a course of treatment, the damage is identical whether or not a compliance report ever got filed. All Colorado did was change where that fight happens, from a regulator's desk to a courtroom, from prevention to autopsy. The discrimination still lands. It just lands on the person first and gets argued about later.

And here's exactly why SB 315 sailed through and got signed by the Governor while Colorado's law got sued into the ground. SB 315 was built to survive all of this. It speaks fluent federal: catastrophic risk, chemical weapons, cyberattack, national security. It even tucked an off-ramp inside itself, a polite little door marked "we will step aside the moment you build a national standard." It is the bill the feds will tolerate. The Colorado-style laws, the ones that actually shield the kid whose resume gets binned and the patient whose pain gets scored down, are the bills the feds take to court.

Which means Illinois didn't just choose the photogenic risk because it photographs well. It chose the risk that comes with federal air cover. The present harm didn't just get left out because it's quiet. No, it got left out because, right now, protecting you from it is the thing that gets a state sued.

Just don't mistake a law about the future for protection in the present. "Catastrophic" is also the word for what is already happening, slowly, legally, profitably, to people who will never trend.

And no law that starts counting at 51 was ever really built to protect the ones who come before.