Apply the Laws You Don't Have

Jensen Huang was asked about the AI executives who keep calling for regulation, and he did not waste the opportunity. They have ulterior reasons, he told CBS. "They're asking to be relieved of the laws we do have." He did not name the executives. He did not name the laws. "We don't need more regulations," he added. "We need to apply the current regulations we have." Hours later, the think tank ITIF published Part 7 of a series called *Mobilizing for Techno-Economic War*, and its central request was a five-year moratorium on state AI laws. Part 7 of anything is not a warning shot. It's a settled position.

The detail matters more than the slogan. ITIF wants Congress to preempt — its word — state AI legislation and state children's online safety laws, and hold both off for five years while it "works toward" national frameworks. Read that again: the children's online safety laws are in there. Nobody has yet explained how a child's exposure to a recommender system threatens American competitiveness. The only way that clause makes sense is as an inventory item — every state statute that has ever annoyed a platform, filed under one heading, frozen with one motion, followed in five years by a national framework drafted in consultation with the parties who asked for the freeze. This is not a plan to legislate later. It's a plan to legislate never and call it scheduling.

I have written this column before, and I resent having to write it again. Last week the lesson of OpenAI's Washington week was that stringency never mattered — only which legislature held the pen. The company funded a super PAC to preempt the states while its president told California to toughen its own rules, and both positions were held sincerely, in the manner of a man holding two umbrellas in a rainstorm. The update is not philosophical. It's that somebody has now written down the part where nobody holds the pen for half a decade. The noun is moratorium. The verb is abolition, with better manners.

Apply the laws we do have, then. Which ones? America has no federal AI statute. The OECD surveyed 6,047 employers across six countries, four of them European, and found that the AI Act — the continent's great regulatory achievement — may not reach most workplace surveillance, and that most of the offending software isn't AI at all. So Huang's prescription amounts to enforcing the law that doesn't exist while freezing the law that does. He is right that his peers have ulterior motives. So does a man selling accelerators into a buildout that only continues if the questions keep being postponed. Correct diagnosis, useful alibi.

The law that does exist, for now, is in Illinois. In July Governor JB Pritzker signed a statute requiring AI companies to conduct annual third-party audits of their safety plans and to disclose incidents to the state. This week he compared the technology to nuclear weapons and asked Congress to do federally what Springfield did at home, while pointedly declining to back Bernie Sanders's superintelligence ban. Audits and disclosure. That is the entire ask — the least you can demand of an industry that has spent two years publishing its own safety frameworks. A five-year freeze on state AI law kills precisely that. Note which direction the preemption always runs.

It would be easier to take "apply the laws we have" seriously if the industry applied anything. PitchBook's Harrison Rolfes, surveying the recent crop of safety announcements and slowdown calls, reached for the word theater. The independent evaluators already exist — METR, Redwood Research, Apollo Research — and a pre-release testing agreement already covers five frontier labs, which tells you the access is achievable and simply isn't being demanded hard enough. Meanwhile Fortune reports that agents broke into Hugging Face in under five days, and the Guardian reports that the Trump family and its allies are getting rich while the White House shields the gold rush. The paper is careful to say the two merely run in parallel, which is the sort of care one takes when describing a coincidence this profitable.

So here is the test, and it is Huang's own. If the current laws are sufficient, name one you would enforce, and against whom. Publish the incident log. Fund the evaluators with the power to delay a release rather than merely to write a report about one. Put the Illinois audit model into federal statute before the freeze, not after it. What we will get instead is the freeze itself: five years of quiet in which the only rule governing frontier AI is the release calendar, which no legislature has ever voted on and no regulator has ever read.

Fifty state legislatures fumbling toward something are worth more than one federal vacuum described as a framework. And if you want to know what is happening to the workforce meanwhile, IBM's Institute for Business Value has just surveyed 8,800 employees: 60 per cent say AI is eroding their skills, and the skill most often cited as declining is critical thinking. The CHROs have responded with a study. Congress, admirably, has responded by promising not to legislate for five years. The only thing being eroded efficiently is the legislature.