New York lawmakers want to make violations of artificial intelligence safety rules considerably more expensive. A bill known as RAISE Act 2.0 would charge frontier AI developers, the companies that build the most advanced AI models, by the day for each violation.
Under Senate Bill S10731, which state Sen. Andrew Gounardes filed Oct. 28, a developer could face fines of up to $1 million per day for a first violation and $3 million per day for later ones. The original Responsible AI Safety and Education (RAISE) Act caps fines at $1 million for a first violation and $3 million for each one after that, no matter how long it lasts.
At $3 million per day, a repeat violation that runs 30 days could cost up to $90 million, compared with $3 million under the original law. The amounts are ceilings, not automatic fines. New York’s attorney general would have to sue a developer to collect them.
The original law takes effect Jan. 1, 2027.
Daily Fines Reach Any Breach of the RAISE Act
Any failure to follow the law, a rule the state writes under it or a court order would draw the daily fine, the bill says. The law as passed fines only four failures, which are not publishing a required safety document, making a false statement, not reporting an incident and not following the company’s own safety plan.
The new bill also adds new duties that could be violated. Each covered developer would hire an outside examiner to check every year that it follows its own safety plan. It would report every three months on how much of its research AI performs. A senior officer would certify its filings under penalty of perjury. A new state office could also limit the computing power large developers use to train models, PYMNTS reported Oct. 30.
The law as passed applies to developers with more than $500 million in annual revenue. Under the new bill, it would apply to any developer that spent more than $1 billion developing frontier models over three years. The revenue test “can miss heavily funded labs with little or no revenue,” the sponsor’s memo says.
Gounardes tied the bill to incidents this year in which AI agents escaped their test environments. In July, agents OpenAI was testing broke out of their test environment and hacked the AI platform Hugging Face, the memo says. “My bill creates safety and transparency requirements with real teeth,” Gounardes said in the announcement.
The bill would give those enterprises an outside check. Each developer’s examiner would share its reports with the state, and the state would make them public.
The bill would regulate the developers. Enterprises have their own limits for AI agents. Eighty-nine percent of enterprises use agentic AI, but 78% would limit agents to reading company directories, PYMNTS Intelligence found in its November report “The PYMNTS Agentic Edge Index: How Leading Companies Put AI to Work.” Just 20% would let agents complete tasks with approval or act on their own.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
