Regulating the Frontier: How California, New York, and Illinois Are Governing the Most Powerful AI Models | American Enterprise Institute - AEI

California, New York and Illinois have now enacted legislation to oversee frontier AI, covering transparency, safety frameworks and incident reporting. The American Enterprise Institute has read all three statutes and pronounced them a sensible beginning, which is the think-tank equivalent of a doctor telling you the tumour is a *starting point*. The frontier, meanwhile, has already packed a bag and moved on to the next jurisdiction, because that is what frontiers do while legislatures are in recess. Three states, three frameworks, one PDF.

The report's central claim is that mandated transparency is the right starting point — holding developers to their safety commitments "without putting a regulator between a model and its release". Read that twice, then read it a third time, because it is doing an enormous amount of quiet work. The entire apparatus of oversight, it turns out, can be delivered by asking the labs to keep being honest and to file the paperwork afterwards. We watched California's Senate Privacy, Digital Technologies and Consumer Protection Committee sit for nearly three hours taking testimony on exactly this territory; the product appears to be a transparency requirement and a warm feeling.

Note also the geography. California, New York and Illinois — the three states containing, between them, most of the labs, most of the money and most of the journalists likely to notice. Everyone else gets to import frontier-AI policy from a Washington think tank with a footnote attached. And "incident reporting" is the loveliest phrase in the whole document: it means the catastrophe gets logged, in triplicate, after the fact, by the company that built the thing. No regulator between the model and its release, you see. Just a form. There's always a form.