AI Frontier Institute

Relates to independent safety evaluations, monitoring, reporting, security, mitigation, employee protections and enforcement relating to frontier artificial intelligence models.

View primary source →

New York S. 10701 (the "TERMINATOR Act") has not been enacted. It was referred to the Rules Committee on 2026-09-18, and the record then shows it recommitted with the enacting clause stricken. If enacted, it would amend the general business law, building on the frontier AI law added by chapter 96 of the laws of 2026. Key provisions: (1) Independent evaluations (new § 1430): a large frontier developer must have an accredited, independent safety evaluator test a frontier model before first deploying it in New York. Models already deployed would need evaluation within 180 days of the effective date. Evaluations must cover at least chemical, biological, radiological and nuclear weapons assistance, offensive cyber capability, autonomous replication, evasion of controls, deception, self-improvement, and safeguard resilience. The developer pays the costs. Evaluators cannot certify a model as safe. The office would set evaluator standards, may reject or revoke accreditation, and may accept substantially equivalent federal evaluations. (2) Reevaluation (§ 1431): another evaluation is needed before deploying a material modification, and after certain new capability discoveries or safety incidents. (3) Monitoring (§ 1432): risk-based catastrophic-risk monitoring after deployment, with data minimization. It does not require general logging of ordinary user prompts. (4) Records (§ 1433): tamper-evident safety records kept at least five years and available to the office or attorney general. (5) Access and weights (§ 1434): security programs for privileged model access, and an independent evaluation of the consequences before releasing or transferring model weights. (6) Mitigation (§ 1435): developers must mitigate material and unreasonable catastrophic risk. The office may require added evaluation, monitoring or safeguards. It may also temporarily restrict activities, in 30-day renewable periods with attorney general review after 60 days, and refer cases to the attorney general for court injunctions. (7) Whistleblower protections (§ 1436): bans retaliation for protected safety disclosures, voids contracts that restrict such disclosures, creates a confidential reporting channel, and provides attorney general enforcement but no private right of action. It also protects authorized good-faith safety research. Amendments to existing sections: transparency reports must summarize independent evaluations (§ 1421(3)(b)). Developers must report "significant safety incidents" within seven days and include specified details (§ 1422, new subdivisions 3-a and 3-b). The act adds civil penalties of up to 0.5% or 1% of annual gross revenue for knowing or willful concealment, interference or retaliation (§ 1427, new subdivisions 4–8). Developers are assessed pro rata for oversight costs (§ 1428(4)). The act expands rulemaking authority (§ 1429). It takes effect one year after becoming law, or one year after chapter 96 of 2026 takes effect if that is later.

← Back to the tracker