One year after passing the Responsible AI Safety and Education Act, known as the RAISE Act, New York lawmakers want to go further. State Senator Andrew Gounardes of Brooklyn and Assemblymember Jordan Wright of Harlem introduced a bill on Thursday, October 29, 2026 that would add new incident disclosure requirements, stronger enforcement powers and independent safety oversight for the largest artificial intelligence developers.
The proposal builds on a law that already requires large developers to publish safety and security protocols and to disclose serious incidents. The new bill would move New York from a disclosure regime toward something closer to financial supervision, in which outside examiners audit compliance and regulators can intervene before harm occurs.
What the RAISE Act Already Does
The RAISE Act, passed in 2025 and signed by Governor Kathy Hochul in December of that year, targets developers of frontier AI models, systems trained with exceptionally high levels of computing power. Covered developers must document safety protocols, assess how their systems could cause critical harm, implement mitigation safeguards and publish disclosures describing those efforts.
The original law set a rapid incident-reporting standard: developers must notify New York regulators within 72 hours of determining that a critical safety incident has occurred. That timeline is significantly shorter than the 15-day window in California's frontier AI framework, which served as a template for New York's approach. The RAISE Act takes effect on January 1, 2027.
Quarterly Reporting on Recursive Self-Improvement
The expansion bill would require large developers to report quarterly on how much of their AI research is automated, a process known as recursive self-improvement, and how they use their computing power. Recursive self-improvement describes AI systems that help design, train or improve successor systems, potentially accelerating capability gains faster than human-only research would.
The 72-Hour Risk-Jump Disclosure
The bill would also require developers to disclose any substantial jump in risk, or in their models' ability to automate AI research, within 72 hours. This is distinct from reporting an incident after it has happened. A risk jump could include a model gaining a dangerous capability during training, a safety evaluation failing unexpectedly, or internal evidence that a system can substantially accelerate its own improvement.
Independent Frontier Examiners
The biggest proposed change is independent auditing. The RAISE Act currently requires developers to disclose their safety plans. The new bill would require developers to retain an independent, state-licensed "frontier examiner" to conduct annual audits of compliance with their safety frameworks, perform broader safety evaluations and make catastrophic risk assessments every 90 days.
PYMNTS notes that this model resembles bank supervision, in which outside auditors evaluate governance, operational risk, controls and remediation. The comparison is deliberate. Banks are examined because their failure can harm people who never chose to be customers. Frontier AI developers, the bill's sponsors argue, pose a similar kind of systemic risk, so self-certification is no longer enough.
DIGIT Compute Powers and Court-Halt Authority
The bill would empower the newly created state Office of Digital Innovation, Governance, Integrity and Trust, known as DIGIT, to limit the computing power large developers devote to training new models or to automated AI research, in order to reduce the risk of catastrophic damage and to prevent a critical safety incident.
It would also empower the Attorney General to ask a court to halt dangerous activity, including by restricting a model's deployment in New York. Compute limits are a powerful lever because training a frontier model requires enormous, concentrated computing resources that can, in principle, be measured and capped. A court-halt power adds a judicial check, so a restriction on deployment would be tested before a judge rather than imposed by an agency alone.
The bill is reported to cover companies that spend more than $1 billion on AI development over three years, a threshold that focuses the regime on the largest developers rather than startups and academic researchers.
Why It Matters for Founders and Investors
For most early-stage founders, this bill does not apply directly, because its thresholds target the largest developers. Its indirect effects, however, could be large. If frontier examiners, quarterly reporting and compute limits become the norm in New York, large model providers may pass compliance costs and delays down the supply chain, affecting startups that build on frontier models.
For investors, the bill is a reminder that AI regulation in the United States is increasingly state-led. With no comprehensive federal framework enacted, and with the federal administration favouring lighter oversight, states are filling the gap in different ways. A startup selling nationally must therefore plan for the strictest state it operates in, not the most permissive.
The bill has only been introduced, not enacted, so its final form may change. But its direction is clear: New York's sponsors describe the RAISE Act as a floor, not a ceiling.
PYMNTS, "New York Bill Would Raise AI Safety Regulations," October 30, 2026
Source: dailyaibrief.com
Daily AI Brief, "New York Bill Would Let State Limit Compute for Risky AI Models," October 29, 2026
Source: legal.io
Legal.io, "New York's RAISE Act Signals a Push Toward State-Level AI Safety Standards"
Read Next
Open Weights Are Not Open Business Models: Mistral Large 4 and the Licensing Trap
Courts Are Writing AI's Rules Now: Three Rulings Founders Should Read
Delhi High Court Orders 24-Hour Takedown of AI Deepfake Videos Targeting Chief Minister
EU Court Adviser Says AI Training Is Legal — but Gemini's Answers May Still Infringe Copyright
AI Regulation After the Midterms: What Founders Should Do Before the January Bills Arrive
DOJ Launches Super Intelligence Task Force to Police AI Misuse



