PRIA Logo™
New YorkS 88282025-2026 Regular SessionSenateWALLET

Relates to transparency and safety requirements for developers of artificial intelligence models

Sponsored By: Andrew Gounardes (Democratic)

Became Law

RULESWAYS AND MEANS

Your PRIA Score

Score Hidden

Personalized for You

How does this bill affect your finances?

Sign up for a PRIA Policy Scan to see your personalized alignment score for this bill and every other piece of legislation we track. We analyze your financial profile against policy provisions to show you exactly what matters to your wallet.

Free to start

Bill Overview

Analyzed Economic Effects

7 provisions identified: 0 benefits, 4 costs, 3 mixed.

Large developers must publish safety plan

Beginning January 1, 2027, large frontier developers must write, follow, and post a Frontier AI Framework on their website. It must cover standards, catastrophic-risk thresholds, mitigations, third‑party checks, securing unreleased model weights, incident response, and internal-use rules. Before or at deployment of a new or majorly updated frontier model, they must post a transparency report with model details; large firms must add risk assessment summaries and third‑party roles. They must review the framework yearly and post material changes with an explanation within 30 days. They may redact sensitive details but must explain the redactions and keep full records for five years.

Register, pay oversight fees, face fines

Starting January 1, 2027, large frontier developers cannot develop, deploy, or run a frontier model in New York without a current disclosure and paying their assessed share. Renew every two years and update after ownership changes or material changes. Disclosures must list all business names, main and New York addresses, owners (5%+ for private or closely held firms; 50%+ for ultimate parents of public companies), and three contacts. The office publishes a list of filers but not their contact details. The office charges pro rata assessments to fund oversight. After notice and a hearing, it can fine $1,000 per day for missing or false disclosures and collect unpaid assessments. The Attorney General can sue for failures to publish or transmit, false statements, missed incident reports, or not following the framework, with penalties up to $1,000,000 for a first violation and up to $3,000,000 for each later one. There is no private right to sue under this article.

New state office for AI safety

The law creates an office in the Department of Financial Services on January 1, 2027. The office runs filings and reports, sets submission tools, and can make rules. It can add reporting like post-incident follow-ups, sharing plans, or require sending safety frameworks. Starting January 1, 2028, it publishes an anonymized yearly report to state leaders, excluding sensitive information.

Report AI safety incidents fast and confidentially

By January 1, 2027, the office sets a public system to report critical safety incidents. Developers must report within 72 hours of determining, or reasonably believing, an incident occurred. If there is an imminent risk of death or serious injury, they must report within 24 hours to the right authority. Large developers must send confidential quarterly summaries of internal catastrophic-risk assessments. Reports to the office are confidential and exempt from public records. A developer can follow an approved federal reporting standard; it must send the same reports to the office at the same time. The office can revoke this approval.

Who New York AI rules cover

Starting January 1, 2027, the law covers “frontier models” trained with more than 10^26 operations. It covers “large frontier developers” with over $500 million in revenue in the prior year. It applies when a model is developed, deployed, or run in whole or in part in New York. Accredited New York colleges and universities doing academic AI research, the Empire AI consortium, and the named institute are exempt. These duties add to, and do not replace, other legal obligations.

No misleading claims about AI safety

Developers may not make materially false or misleading claims about catastrophic risks or their risk management. Large developers may not misstate how they follow their published framework. Good‑faith, reasonable statements are allowed. This applies beginning January 1, 2027.

Stock price drops not recoverable here

A drop in equity value does not count as property damage under this law. Investors cannot recover stock or equity value declines as damages under this article. This applies starting January 1, 2027.

Sponsors & Cosponsors

Sponsor

  • Andrew Gounardes

    Democratic • Senate

Cosponsors

There are no cosponsors for this bill.

Roll Call Votes

All Roll Calls

Yes: 77 • No: 1

Senate vote • 1/28/2026

FLOOR Vote

Yes: 58 • No: 1

committee vote • 1/20/2026

Rules Committee Vote

Yes: 19 • No: 0

Actions Timeline

  1. SIGNED CHAP.96

    3/27/2026Senate

  2. DELIVERED TO GOVERNOR

    3/20/2026Senate

  3. RETURNED TO SENATE

    3/11/2026House

  4. PASSED ASSEMBLY

    3/11/2026House

  5. ORDERED TO THIRD READING RULES CAL.85

    2/25/2026House

  6. SUBSTITUTED FOR A9449

    2/25/2026House

  7. REFERRED TO WAYS AND MEANS

    1/28/2026House

  8. DELIVERED TO ASSEMBLY

    1/28/2026Senate

  9. PASSED SENATE

    1/28/2026Senate

  10. ORDERED TO THIRD READING CAL.94

    1/20/2026Senate

  11. REFERRED TO RULES

    1/8/2026Senate

Bill Text

  • Original

    1/8/2026

    PDF

Related Bills

Back to State Legislation