Sen. Gounardes Proposes Major New AI Safety and Transparency Regulations

Albany has produced a bill with a kill switch in it, courtesy of State Senator Andrew Gounardes and Assemblymember Jordan Wright, and in a legislature that measures urgency in decades this counts as a sprint. Not a task force, not a set of principles, not a listening tour — a switch, which somebody will presumably be permitted to press. Who that somebody is, what precisely counts as "catastrophic risk" and whether the button comes with a little perspex cover are details the press release declines to share. But the word "kill" is now in a New York bill, and that is the sentence the sponsors will be repeating on cable.

It builds on Gounardes' Responsible AI Safety and Education Act, which is called RAISE, because the first act of every AI legislator is to reverse-engineer an acronym and the second is to name the sequel after it. The substance here is independent third-party analysis and assessments of existing models — a genuinely decent idea, which immediately raises the question of who pays the auditor. The firms with the logs, the lawyers and the lobbying budget are the same firms being assessed. You may be spotting the commercial tension. RAISE. Somewhere a consultant is very proud and doesn't know why.

Meanwhile California held the committee hearing we noted at which executives were questioned and comprehensive safeguards were promised; New York has answered with a switch, which is more than Washington has managed, which is faint praise measured against a void. And "advanced AI" remains undefined in the release, so the bill currently regulates a vibe. Still, someone in Albany looked at a model capable of catastrophic risk and asked, reasonably, whether there ought to be an off switch. Wait for the amendments, in which the switch is renamed a "dynamic risk-mitigation interface" and leased back to the industry. A senator, an assemblymember, a press release and no bill text — the full apparatus of American AI governance.