What's happening
This bill requires independent safety evaluations, monitoring, reporting, and protections for advanced AI models in New York.
Why it matters
Creates a comprehensive state-level oversight structure for "large frontier developers" of advanced AI in New York. , requiring independent safety evaluations and ongoing monitoring before and after deploying powerful AI models—this directly impacts major tech companies, AI startups, and research institutions operating or offering services in New York, setting a high bar compared to most current state or federal AI regulations.
Empowers a state office to intervene rapidly if an AI model poses a material and unreasonable risk of catastrophic harm (such as aiding in the development of chemical, biological, or nuclear weapons, or enabling dangerous autonomous behavior). the office can temporarily restrict deployment or access, and the Attorney General can seek court injunctions, making New York one of the first states with specific legal mechanisms to halt risky AI activity.
Institutes robust whistleblower and security protections. , including anti-retaliation rules for employees who report safety concerns, mandatory tamper-evident recordkeeping, and civil penalties (up to 1% of annual gross revenue) for companies that conceal incidents or interfere with independent evaluations—this aims to create accountability and transparency, which could serve as a model for national or global AI governance.
Albany, before your first meeting.
Every bill that moved yesterday, summarized in a sentence and ranked by impact. The Albany Scorecard follows on Sundays.
Free. Unsubscribe in one click. Want alerts on specific bills? Create an account.
Who cares
- Watching
Large Frontier AI Developers
Must comply with new requirements for independent safety evaluations, post-deployment monitoring, reporting incidents, mitigation, record-keeping, cybersecurity standards, and employee protections. They face new costs, operational constraints, and significant penalties for violations.
AI Safety Evaluators & Technical Experts
Gain new roles and opportunities as accredited independent evaluators, with formalized standards, oversight, and potentially increased demand for their services.
State Regulators and the Attorney General
Receive expanded authority and responsibilities to oversee, enforce, and regulate AI safety, including rulemaking, investigations, contracting with experts, and imposing penalties or restrictions on risky AI deployments.
AI System Employees, Contractors, and Whistleblowers
Receive new legal protections against retaliation for reporting safety risks or violations, with mechanisms for confidential complaints and remedies for retaliation.
Public and Civil Society Organizations
Stand to benefit from enhanced oversight, transparency, and mitigation of catastrophic AI risks, as well as new channels for ensuring AI systems do not threaten public safety or welfare.
Headline styles
Regulate AI Safety Oversight
Plain English, a few words. Used as the page title.
Sponsors
No co-sponsors yet.
History
- Recommit, enacting clause stricken
- Referred to rules
Full text
11 pagesUnderlined is new[Struck] is removedOfficial text on nysenate.gov ↗
STATE OF NEW YORK
10701
IN SENATE
September 18, 2026
Introduced by Sen. FAHY -- read twice and ordered printed, and when printed to be committed to the Committee on Rules
AN ACT to amend the general business law, in relation to independent safety evaluations, monitoring, reporting, security, mitigation, employee protections and enforcement relating to frontier artificial intelligence models
The People of the State of New York, represented in Senate and Assembly, do enact as follows:
Section 1. This act shall be known and may be cited as the "technical evaluation, risk monitoring, incident notification, AI testing, oversight, and response act" or the "TERMINATOR Act".
§ 2. Section 1420 of the general business law is amended by adding seven new subdivisions 17, 18, 19, 20, 21, 22 and 23 to read as follows:
17. "Independent safety evaluator" means a person or entity that:
(a) possesses technical expertise and experience appropriate to the evaluation of frontier artificial intelligence models and catastrophic risk;
(b) is not an affiliate of the large frontier developer whose frontier model is being evaluated;
(c) has no material financial interest in such large frontier developer other than compensation for services provided in connection with an evaluation required by this article; and
(d) is accredited or otherwise recognized by the office pursuant to standards established by the office.
18. "Material modification" means a modification to a frontier model or the manner in which such model is deployed that could reasonably be expected to materially increase a capability relevant to catastrophic risk or materially reduce the effectiveness of a safeguard designed to prevent or mitigate catastrophic risk.
19. "Significant safety incident" means an event involving a frontier model that materially increases catastrophic risk or provides credible evidence of a capability, vulnerability, failure, circumvention, or compromise of a safety, security, or control mechanism that could materially increase catastrophic risk. A significant safety incident need
EXPLANATION--Matter in italics (underscored) is new; matter in brackets
[ ] is old law to be omitted.
LBD16379-01-6
not result in death, bodily injury, property damage, or the materialization of catastrophic risk.
20. "Privileged frontier model access" means access to a frontier model that materially exceeds access generally provided to users of such model and that could materially increase catastrophic risk, including access to model weights, internal model checkpoints, a version of a model with safeguards disabled or materially reduced, privileged finetuning capabilities, or elevated authority to access external tools, systems, credentials, or computing resources.
21. "Risk-based catastrophic risk monitoring" means a program reasonably designed, in light of the capabilities, deployment context, and reasonably foreseeable uses of a frontier model, to detect significant safety incidents, material changes in capabilities relevant to catastrophic risk, and material failures or circumvention of safeguards designed to prevent or mitigate catastrophic risk.
22. "Tamper-evident" means maintained using technical controls reasonably designed to permit detection of unauthorized alteration or deletion.
23. "Catastrophic harm" means death, serious injury, property damage, or loss of property of the nature and magnitude described in subdivision three of this section arising from conduct described in such subdivision.
§ 3. Paragraph (b) of subdivision 3 of section 1421 of the general business law, as added by chapter 96 of the laws of 2026, is amended to read as follows:
(b) Before, or concurrently with, deploying a new frontier model or a substantially modified version of an existing frontier model, a large frontier developer shall include in the transparency report required by paragraph (a) of this subdivision, summaries of all of the following:
(i) assessments of catastrophic risks from the frontier model conducted pursuant to the large frontier developer's frontier AI framework;
(ii) the results of the assessments under subparagraph (i) of this paragraph;
(iii) the extent to which third-party evaluators were involved; [and]
(iv) other steps taken to fulfill the requirements of the frontier AI framework with respect to the frontier model; and
(v) a summary of the material findings, limitations, uncertainties, identified capabilities, safeguard performance, and identified catastrophic risks contained in any independent safety evaluation required pursuant to section fourteen hundred thirty or fourteen hundred thirty-one of this article. Nothing in this subparagraph shall require publication of information that may be redacted pursuant to subdivision five of this section.
§ 4. Subdivision 2 of section 1422 of the general business law is amended by adding a new paragraph (c), and three new subdivisions 3-a, 3-b and 11 are added to read as follows:
(c) A large frontier developer shall, together with each summary of an assessment of catastrophic risk transmitted pursuant to this subdivision, provide, to the extent reasonably available, technical information reasonably necessary for the office to evaluate the basis for such assessment, including, as applicable, relevant evaluation results, monitoring outputs, model outputs or trajectories, safety or security testing results, and other technical information supporting such assessment.
3-a. (a) A large frontier developer shall report a significant safety incident pertaining to one or more of its frontier models to the office as soon as practicable but in no event later than seven days after determining that such incident has occurred or learning facts sufficient to establish a reasonable belief that such incident has occurred.
(b) Nothing in this subdivision shall alter a more expeditious reporting requirement applicable to a critical safety incident pursuant to subdivision three of this section.
(c) Information submitted pursuant to this subdivision shall be subject to the same confidentiality, disclosure, and information-sharing protections applicable to critical safety incident reports under this section.
(d) No large frontier developer shall knowingly structure or administer its monitoring, incident-review, investigation, or recordkeeping practices for the purpose of avoiding the detection or reporting of a significant safety incident or critical safety incident required to be reported under this article.
3-b. Any report by a large frontier developer of a significant safety incident required pursuant to subdivision three-a of this section or a critical safety incident required pursuant to subdivision three of this section shall include, to the extent reasonably available:
(a) the date and nature of the incident;
(b) the frontier model or models involved;
(c) the capability, vulnerability, safety mechanism, security mechanism, or control implicated by the incident;
(d) the large frontier developer's assessment of the actual or potential consequences of the incident;
(e) corrective or mitigating action taken or planned by the large frontier developer; and
(f) technical information reasonably necessary for the office to assess the nature, cause, and significance of the incident, including, where applicable, relevant logs, model outputs, agent trajectories, tool calls, monitoring outputs, security alerts, and evaluation results.
11. Notwithstanding subdivisions eight and nine of this section, compliance with a federal law, regulation, or guidance document designated by the office as satisfying a reporting requirement under this section shall not relieve a large frontier developer of the obligation to provide the information required pursuant to subdivision three-b of this section, except to the extent that substantially equivalent information has been transmitted to the office pursuant to such federal reporting regime.
§ 5. The general business law is amended by adding seven new sections 1430, 1431, 1432, 1433, 1434, 1435 and 1436 to read as follows:
§ 1430. Independent frontier model safety evaluations. 1. A large frontier developer shall, before first deploying a frontier model in whole or in part in this state, cause an independent safety evaluator to conduct a technical evaluation of such model's capabilities, safeguards, and potential to materially contribute to catastrophic risk.
2. An evaluation required pursuant to this section shall include technical testing, including adversarial testing where appropriate, and shall separately assess where applicable:
(a) capabilities of the underlying frontier model;
(b) capabilities reasonably accessible under the intended deployment configuration; and
(c) the effectiveness and circumvention resistance of safeguards separating underlying capabilities from those reasonably accessible under the intended deployment configuration.
3. An evaluation shall assess, at a minimum:
(a) capabilities that could materially assist in the development, production, or release of chemical, biological, radiological, or nuclear weapons;
(b) offensive cyber capabilities that could materially contribute to catastrophic risk;
(c) capacity to autonomously replicate, propagate, persist, or acquire computing resources or credentials;
(d) capacity to evade, circumvent, or subvert material controls imposed by a developer or user;
(e) capacity to engage in material deceptive behavior, conceal material capabilities, or manipulate or evade safety evaluations or monitoring;
(f) capacity to materially assist in its own development or improvement, or the development or improvement of another frontier model, in a manner that could materially increase catastrophic risk;
(g) effectiveness and resilience of safeguards designed to prevent or mitigate catastrophic risk, including susceptibility of such safeguards to circumvention; and
(h) other capabilities or safeguards relevant to catastrophic risk as the office may designate by regulation.
The requirements of paragraphs (a) through (g) of this subdivision constitute minimum requirements and may not be waived or eliminated by regulation.
4. A large frontier developer shall provide the independent safety evaluator access reasonably necessary to conduct a meaningful technical evaluation, including, where reasonably necessary, access to relevant model checkpoints, interfaces, documentation, testing environments, fine-tuning capabilities, and versions or configurations of the model in which safeguards are disabled or materially reduced.
5. The independent safety evaluator shall prepare a written report describing the methodology, material findings, limitations, uncertainties, identified capabilities, safeguard performance, and identified catastrophic risks. An unredacted copy shall be transmitted to the office.
6. A large frontier developer shall pay the reasonable costs of an evaluation required pursuant to this section. Compensation of an independent safety evaluator shall not be contingent upon the results or conclusions of an evaluation.
7. Nothing in this section shall require or authorize an independent safety evaluator to certify that a frontier model is safe or free from catastrophic risk.
8. (a) The office shall establish by regulation minimum qualifications, technical competency, independence, conflict-of-interest, professional conduct and ethical standards, cybersecurity, confidentiality, personnel-access, information-storage and destruction, and incident-reporting requirements for independent safety evaluators.
(b) Such regulations shall establish standards reasonably designed to ensure the independence, integrity, and professional judgment of independent safety evaluators, including standards governing financial interests, financial dependence, prior or concurrent professional relationships, compensation arrangements, and other circumstances that could reasonably be expected to impair an evaluator's independent professional judgment.
(c) Such regulations shall provide a means for temporary recognition of otherwise qualified evaluators where necessary to ensure adequate evaluator capacity.
9. The office may reject or revoke the accreditation or recognition of an independent safety evaluator upon a finding that such evaluator lacks the technical competence, independence, security practices, professional integrity, or other qualifications required pursuant to this section, and may require a large frontier developer to retain a different independent safety evaluator.
10. The office may determine that an independent technical safety evaluation conducted pursuant to a federal law, regulation, or other governmental regulatory regime satisfies all or part of the requirements of this section where the office determines that such evaluation is substantially equivalent to or more stringent than the requirements of this article. The office may require supplemental evaluation of any capability, safeguard, or risk required to be assessed under this article but not adequately assessed by such evaluation. An evaluation performed solely pursuant to a developer's internal policy or voluntary safety framework shall not qualify under this subdivision solely because such evaluation was conducted or commissioned by the developer.
11. A frontier model deployed in whole or in part in this state before the effective date of this section shall receive the evaluation required by this section within one hundred eighty days after such effective date.
Reflowed for reading: line numbers and page breaks removed. Cite the official text.
Sponsor's memo
Title of bill
An act to amend the general business law, in relation to independent safety evaluations, monitoring, reporting, security, mitigation, employ- ee protections and enforcement relating to frontier artificial intelli- gence models
Purpose
Enacts the technical evaluation, risk monitoring, incident notification, Al testing, oversight, and response act or the TERMINATOR Act
Summary of provisions
Section 1 establishes the short title, the "TERMINATOR Act."
Read the full memo (973 words)
Section 2 amends General Business Law § 1420 to add definitions relating to independent safety evaluators, material modifications, significant safety incidents, privileged frontier model access, risk-based monitor- ing, tamper-evident records, and catastrophic harm.
Section 3 amends General Business Law § 1421 to require transparency reports to summarize findings from required independent safety evalu- ations.
Section 4 amends General Business Law § 1422 to expand catastrophic-risk and safety-incident reporting, require supporting technical information, and address substantially equivalent federal reporting.
Section 5 adds General Business Law §§ 1430 through 1436 to establish independent safety evaluations, reevaluation following material changes or new risks, post-deployment monitoring, tamper-evident records, privi- leged-access and model-weight security requirements, catastrophic-risk mitigation authority, and protected safety disclosures and research.
Section 6 amends General Business Law § 1427 to establish enhanced civil penalties for concealment, obstruction, interference with independent evaluations, and related misconduct.
Section 7 amends General Business Law § 1428 to authorize existing developer assessments to support administration and enforcement of the expanded requirements.
Section 8 amends General Business Law § 1429 to expand rulemaking, tech- nical-assistance, consultation, and intergovernmental-coordination authority.
Section 9 contains a severability clause. Section 10 establishes the effective date.
Justification
The rapid proliferation of artificial intelligence (Al) continues to raise serious concerns about potentially catastrophic risks associated with the technology. Increasingly, leading Al developers are documenting enhanced offensive cyber capabilities, evidence of oversight-evasion and deception, and growing capacity for autonomous replication. The 2026 International Al Safety Report warns that the rapid expansion of fron- tier capabilities is outpacing the effectiveness of imperfect safe- guards, underscoring the need for ongoing oversight and independent testing of model safety.
New York established an important framework for addressing these risks in the RAISE Act, which, among other requirements, requires large fron- tier developers to establish and follow frontier Al safety frameworks, assess catastrophic risks, protect model weights, and report critical safety incidents. However, New York's framework substantially relies on Al developers' self-assessment of their own models. Developers determine the effectiveness of their own safeguards and are able to characterize the results of those assessments without submitting the underlying data or allowing for independent external review. Compliance with a develop- er's safety framework does not necessarily establish whether an advanced Al model actually possesses dangerous capabilities or whether the safe- guards intended to constrain those capabilities are actually functional. Examining this problem, a 2026 Model Evaluation and Threat Research (METR) pilot was given access to the most capable non-public models of companies, including Anthropic, Google, Meta, and OpenAl. The report recommended that third-party risk assessments become standard industry practice
This legislation establishes an independent technical safety layer on top of the existing RAISE framework both before and after frontier model deployment. Before a covered frontier model could be deployed in New York, an independent evaluator would be required to test the model's capabilities and safeguards, including through adversarial testing where appropriate. Independent testing and requiring submission of underlying data pre-deployment and after safety incidents will allow regulators to more thoroughly examine the safety claims of developers, rather than solely relying on the corporate characterization of those claims.
Moreover, this legislation recognizes that critical safety issues may still exist after deployment. Post-training, fine-tuning, increased autonomy, access to tools or computing resources, and other post-deploy- ment modifications may change a model's capabilities in a manner that materially impacts the model's associated safety risks. Recognizing this hazard, the bill requires the reevaluation of frontier models following material modifications or significant new evidence of danger- ous capabilities. It also provides for risk-based post-deployment moni- toring, reporting of serious warning signs and near misses, and the preservation of safety-relevant records. Where evaluations, monitoring, safety incidents, or other evidence identify an unreasonable risk of catastrophic harm, the legislation establishes an escalating framework for regulatory intervention, balancing the seriousness of risk to the public with allowing frontier developers to mitigate their models' risks. Upon the identification of such a risk, developers must implement mitigation safeguards and may be required to submit to additional moni- toring by the Office. If the risk remains unmitigated, the bill creates a process for temporary and targeted restrictions, subject to notice and periodic renewal. More significant intervention is reserved for findings of substantial evidence of imminent and unreasonable risk of catastroph- ic harm, upon which the bill authorizes the Attorney General to seek injunctive relief in court.
More broadly, the bill also addresses the safety infrastructure ecosys- tem surrounding frontier models. The legislation requires tamper-evident records of safety actions and establishes cybersecurity requirements for privileged access and model weights. Additionally, the bill creates whistleblower protections for employees and contractors who make good- faith safety disclosures. Enhanced penalties are also established for deliberate obstruction of oversight, including safety records tampering.
Importantly, the TERMINATOR Act is not a general regulatory regime for artificial intelligence, and operates within the existing RAISE frame- work. It is applicable to a limited class of large frontier developers, and its requirements are directed specifically at catastrophic risk. This act creates meaningful independent oversight of the most consequen- tial risks associated with frontier models, while avoiding regulation of ordinary Al systems.
Fiscal implications
To be determined.
Effective date
This act shall take effect one year after it shall have become law; provided, however, that if chapter 96 of the laws of 2026 shall not have taken effect on or before such date, then this act shall take effect one year after such chapter takes effect.
Summaries, headlines and scores are AI-generated from the official bill text. How we score bills. Source: NYS Open Legislation.