Self-regulation is the AI industry’s default, and New York just decided it isn’t good enough. Governor Kathy Hochul launched what she calls the first state-run AI testing program in the US. Her pitch: “The stakes are too high to simply take AI companies at their word.”
The work will happen at the New York Center for AI Responsibility and Research at Binghamton University. More than 80 state scientists will contribute, running on compute from Empire AI, the state’s $500 million public-private partnership. The center will test advanced systems, study their risks, and hand evidence to public officials. It will also look at responsible uses in areas like medicine and scientific research.
Why does this matter? Right now, most safety evaluation of frontier models comes from the labs themselves. OpenAI, Anthropic and Google publish system cards and run red-team exercises, but they grade their own homework. Third-party groups and government bodies like the UK’s AI Security Institute and the US federal AI safety institute have done some outside testing, but Washington has largely stepped back. New York is filling that vacuum, much as California has done with its own AI bills.
There’s a catch. The center can only study what companies make public. Unreleased models stay out of reach, and that’s where the riskiest capabilities usually sit.
The RAISE Act, which Hochul signed last year, partly addresses this. Large AI developers must register with the state by January 1, 2027. After that, they have to publish safety frameworks and transparency reports, and report critical safety incidents within 72 hours. It won’t give testers access to unreleased models, but it forces more disclosure.
Will it change how labs behave? Probably not overnight. But a credible, state-backed testing body gives regulators and buyers something they haven’t had: evidence that doesn’t come from the vendor. Other states will be watching, and if the results are useful, expect copycats.
