Jump to content

Artificial Intelligence Risk Management and Security Act of 2026

From AI Law Wiki

AI-generated text. This page was generated using artificial intelligence.

Law & policy information
Name
Artificial Intelligence Risk Management and Security Act of 2026
Jurisdiction
United States (federal)
Instrument type
Proposed legislation
Bill / legal citation
S. 5576, 119th Congress.[1]
Legislative or adoption status
Formally introduced September 29, 2026 and referred to Senate Commerce, Science, and Transportation.[1] Sponsors announced the proposal September 24. On September 29, Senator Warner asked unanimous consent for Senate passage; Senator Ted Cruz objected, so the request did not pass the proposal.[2][3]
Effective dates
Not applicable to a proposal; no enactment or effective date has been verified.
Last source verification
October 15, 2026 — complete September 29 introduced print and current official bill-status record reviewed.[4][1]

The Artificial Intelligence Risk Management and Security Act of 2026 is proposed United States federal legislation announced on September 24, 2026. It has not been enacted.[5]

Proposal

Senators Mark Warner and Brian Schatz announced the Artificial Intelligence Risk Management and Security Act of 2026 proposal on September 24; the official record dates formal introduction to September 29.[1] The proposal would create an AI Safety Board within the Department of Commerce and require developers of frontier models to provide access to the board at least 45 calendar days before introducing a frontier model into interstate or foreign commerce. It would also require model safety plans and establish incident-reporting and testing measures.[6][7][4]

Scope and model access

The introduced bill defines a frontier model by serious-risk capabilities affecting national security, national economic security, or public health or safety, including capabilities that could be obtained by modifying the model. It would require developers to give the Board access at least 45 calendar days before introducing a frontier model into interstate or foreign commerce, including the weights, configuration files, runtimes or software libraries needed to operate it. The Board’s evaluation processes would also address nonpublic development models, variants and checkpoints.[4]

Standards and safety plans

The Commerce Secretary would establish the Board within 30 days after enactment. The Board would submit proposed technical standards within 90 days of its establishment; the Secretary would then adopt standards by rule within 30 days of receiving them, with national-security modifications permitted and reasons published. Developers would have to comply with applicable adopted standards. Violations could incur civil penalties of up to $250,000 each, with each continuing day treated separately; the Attorney General could seek an injunction or recover penalties at the Secretary’s request.[4]

Developers would have to develop, publish, follow and submit model safety plans for systems or models they create, substantially modify, or use to train or evaluate other models. Each plan would identify the covered model or system, assess its capabilities or risks, specify lifecycle mitigation measures and name the responsible corporate officer.[4]

Incident reporting and agent assurance

Section 5 would establish voluntary reporting mechanisms within 180 days after enactment and a public incident database and flaw catalogue within one year. Frontier-model developers or providers, and critical-infrastructure operators using AI in industrial-control or other operational technologies, would have to report confirmed safety or security incidents within 30 days of confirmation, or within 72 hours if an incident poses an imminent threat to national security, critical infrastructure or public safety. Reporting would cover nonpublic models and incidents during development, testing and operation. The bill provides confidentiality, privilege and anonymisation safeguards and consultation with the developer or provider before incident information is published.[4]

Section 6 would require NIST to develop an agentic-AI profile accompanying its AI Risk Management Framework within 18 months after enactment and to initiate work on a common agent-assurance documentation template within the same period. The profile would address autonomy levels and agent-specific risks, including identity, authentication and authorisation; the template would document intended uses, authority boundaries, access, evaluations and limitations.[4]

Status and coverage

The sponsors announced the proposal on September 24. Warner asked unanimous consent to pass it on September 29, but Cruz objected; the bill was not passed by that request. The official congressional record identifies the bill as S. 5576, formally introduced by Warner on September 29 with original cosponsors Brian Schatz and Andy Kim. It was read twice and referred that day to the Committee on Commerce, Science, and Transportation, the latest action recorded as reviewed October 14.[1][8][9] See September 24 news coverage and federal authorities.

References