New York bill can make AI labs get external safety audits
The bill makes large AI developers use a licensed examiner for each annual audit and for risk checks each 90 days.
Senator Gounardes of New York proposed a bill that adds to the RAISE Act, which only makes developers disclose their safety plans. Large developers must report in 72 hours a large increase in risk or in the AI research that their models automate. A new New York office can limit the computing capacity that they use for training.
How it works
A licensed examiner does an audit of each large developer each year. The examiner also does wider safety tests and risk estimates each 90 days. Developers tell the office each three months how much of their AI research is automated.
The new powers
- The New York office DIGIT can limit the computing capacity for training and automated research.
- The Attorney General can tell a court to stop dangerous work.
- A court can limit the use of a model in New York.
What is not known
The source gives no vote date and no cost. Gounardes says that large companies cannot control their own work. The source says that the tech industry opposed the RAISE Act. The Trump administration wants regulators to stay away from AI developers.
Insight: Three rule models compete for AI oversight
In October 2026, Anthropic backed a US bill that lets the government stop a release only at a risk of $1 billion in damage. The New York text, as reported, names no such limit. The US group Americans for Responsible Innovation wants reports to a federal office. This reads as a race to set the model for AI oversight, and Washington is silent. Plan for more than one reporting rule.
Sources
Posted
