Artificial Intelligence Risk Management and Security Act (S. 5576)

Note

Provenance of the text quoted below, stated before the quotations

The document read in full is a PDF published by the sponsoring office. It carries no bill number — its bill line reads S. ll and it bears the drafting reference HLA26981 G44 S.L.C., which marks it as a Senate Legislative Counsel draft posted by the office rather than the bill as printed on introduction.

The legislative record independently confirms that S. 5576 exists, was introduced on 2026-09-29 by Senator Warner with Senators Schatz and Kim as original cosponsors, was read twice and referred to the Committee on Commerce, Science, and Transportation, and bears the official title “A bill to establish the Artificial Intelligence Safety Board, and for other purposes.” — which matches section 3 of the retrieved draft.

It does not confirm that the introduced text is word-for-word the retrieved draft. Every quotation on this page is a quotation from that draft and is attributed to it. A copy of S. 5576 as introduced has not been retrieved.

Why regulates is recorded as both

The following entities shall report any confirmed artificial intelligence safety incident or artificial intelligence security incident within 30 days of confirmation of such incident and within 72 hours if such incident poses an imminent threat to national security, critical infrastructure, or public safety: (i) Any developer or provider of a frontier artificial intelligence model. (ii) Any operator of critical infrastructure that utilizes an artificial intelligence model in the context of managing industrial control systems or other operational technologies.

Limb (i) binds whoever builds or serves the model. Limb (ii) binds whoever runs a deployed system with an artificial intelligence model in the control path. The pre-release and safety-plan duties in sections 3 and 4 of the draft bind developers only. The duty is therefore split, and the entry records both.

Limb (ii) is the first provision in any United States federal instrument in this register to place an obligation on the operator of a deployed system rather than on the party that built the model.

Open-weights applicability

APPLICABILITY.—For the purposes of subparagraph (A), the reporting requirement under this paragraph shall apply regardless of whether the frontier artificial intelligence model, or any model variant, checkpoint, or other version of such model involved in the incident, is or is intended to be made publicly available, and regardless of whether the incident occurs during development, training, testing, evaluation, red-teaming, deployment, or operation.

This is the only provision in this register that states expressly that an obligation applies irrespective of whether model weights are published.

What the reporting duty attaches to

The term ''artificial intelligence safety incident'' means an event that materially increases the risk that operation of an artificial intelligence system leads to a state in which human life, health, property, or the environment is endangered.

Section 2(3) of the draft, which defines “artificial intelligence capabilities or risk assessment”, enumerates at subparagraph (D) the capability “to evade the control of, or act outside the intended instructions of, its developer or operator”.

Section 2(6) defines “artificial intelligence security incident”. The retrieved text was served in truncated fragments at that point and the definition ends mid-word, so it has not been read whole and nothing is asserted about its remainder.

No stop requirement

No obligation to maintain a capability to shut down, disable, halt or suspend a model was identified anywhere in the retrieved draft. The only termination provision located is section 3(a)(6)(A)(iv)(VI) — “conditions, policies, procedures, and technical controls for prompt termination of any evaluation” — which terminates an evaluation, not a system.

A cosponsor is quoted in both sponsoring offices’ releases as follows:

Our bill is about making sure humans remain in control of AI models.

Neither release, and no provision of the retrieved draft, carries a human-control or stop requirement. The requires field above records what the text contains.

Agentic provisions

Section 6(a) of the draft directs the National Institute of Standards and Technology to develop an “Agentic AI Profile” within 18 months as “a companion resource to the Framework” that “shall not modify or supersede the Framework”. Section 6(b) directs a documentation template for artificial intelligence agents. Neither places an obligation on a private party and neither requires a stop capability.