Transparency, Safety Are Focuses for Illinois AI Laws

Ah, splendid. Illinois has decided that the most effective way to govern a technology that could end civilization is to demand a transparency report. The Land of Lincoln defines "catastrophic" frontier AI risk as anything that kills more than 50 people or causes more than $1 million in property damage in a single incident. So 49 deaths and a half-million dollars of damage? That's just an ordinary Tuesday, apparently. And "materially contribute" — a phrase so slippery it could pass as diplomatic. We're all deeply reassured that the state has thought this through with the precision of a law school exam.

But there's more. The same package updates the Illinois Human Rights Act to prohibit employers from using AI that discriminates on protected classes. Because the real problem with AI discrimination was that it lacked proper paperwork. Now the algorithms can simply discriminate on postcode, school tie, and credit score instead — the traditional, legally licensed channels. And let's not pretend the transparency reports will be anything but PDFs filed with some underpaid clerk in Springfield, eliciting a sternly worded letter should a lab ever admit its model might kill 51 people. That'll happen the day after a frontier lab voluntarily admits that its model is a menace, which is to say precisely never.

So, a round of applause for Illinois, which has reduced AI safety to a form-filling exercise. The Human Rights Act update at least has teeth — but no enforcement mechanism, so the AI can discriminate as long as nobody catches it. The "catastrophic" threshold is set so high that any honest assessment will conclude there's no risk whatsoever. After all, the first model to cause a million dollars' damage will probably do it in a server room during a weekend overheating incident, and that can always be written off as depreciation. Meanwhile, the rather more pressing risk — that a model will kill exactly 49 people, say, or quietly grind down a delivery driver's pay through a biased scheduling algorithm — remains entirely legal, because that's not "catastrophic," and the law is all about the big numbers. Catastrophe, averted; careers, unaffected.