The AI Threat Output and Monitoring Incident Containment Act is a bill introduced in the United States House of Representatives on July 27, 2026 by Representative Celeste Maloy (R-UT-2), "for herself and Ms. Jacobs." It would establish an Advanced Artificial Intelligence Nuclear Evaluation Program to test advanced AI systems for nuclear-incident and loss-of-control risks, with mandatory participation by large advanced AI developers. The bill was referred the same day to the Committee on Science, Space, and Technology and remains at introduction (Source: congress.gov). The full bill text is summarized at H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text).
Status and timeline
| Field | Value |
|---|---|
| Bill number | H.R.9965, 119th Congress, 2d Session |
| Introduced | July 27, 2026 |
| Sponsor | Rep. Celeste Maloy (R-UT-2), with Ms. Jacobs |
| Referral | Committee on Science, Space, and Technology (July 27, 2026) |
| Text versions | One — Introduced in House |
| Status | Introduced |
Scope and definitions
The Act reaches advanced artificial intelligence, defined as AI trained using more than 10^26 integer or floating-point operations, or other AI the Secretary of Energy specifies by rule after considering research developments, the need to cover AI that may pose a serious risk of an AI nuclear incident, and data collected through the Program (H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text)). The 10^26 figure matches the threshold used in New York RAISE Act, California SB 53 and New York A8833 — Understanding Artificial Intelligence Act; see Compute Thresholds.
Obligations attach to a large advanced artificial intelligence developer — an entity that develops, owns, or substantially modifies advanced AI for interstate or foreign commerce, and that with its affiliates spent at least $2 billion over the five preceding calendar years on investments to develop or modify AI, counting compute, data procurement, personnel, infrastructure, and research and development. The bill uses an expenditure test rather than the revenue test adopted in the California and New York frontier-model statutes.
The AI nuclear incident the Program tests for is defined in six limbs, each reaching attempts as well as completed conduct: generation of technical information reasonably likely to be a substantial factor in the unlawful development, acquisition, or use of a nuclear weapon or nuclear material; generation of Restricted Data as defined in the Atomic Energy Act of 1954; a loss-of-control scenario involving systems connected to such a weapon, material, a nuclear facility, or the nuclear stockpile; a foreign terrorist organization or foreign adversary obtaining unauthorized access to, manipulating, or interfering with such a system; weaponization by such an organization or adversary; and scheming behavior relating to any of them.
The bill supplies statutory definitions of three terms that otherwise circulate mainly in technical literature. Loss-of-control scenario requires both a described behavior — pursuing an objective different from any intended by a human developer or operator, by acting contrary to instruction in a high-stakes context, altering constraints without authorization, or subverting an oversight or shutdown mechanism — and a quantified harm floor of at least five deaths, serious bodily injury to at least 50 individuals, or more than $100 million in damages. Evaluation awareness is the capability to determine that a test is under way and modify behavior so the result is unrepresentative. Scheming behavior is behavior carried out to deceive a human, including hiding a capability or objective or attempting to subvert an oversight or shutdown mechanism. See AI Scheming, Deceptive Alignment, Unverbalized Evaluation Awareness.
Key provisions
The Secretary of Energy must establish the Program within 90 days of enactment. Its six functions are testing and evaluation for AI nuclear incident likelihood; testing protocols addressing "practical jailbreaking techniques," including red-teaming by entities with expertise the Secretary determines comparable to sophisticated malicious actors; facilitation, to the extent practicable, of independent third-party assessments and blind model evaluations; a report to each participant on its own testing outcomes and the risks and safety measures evaluated; development of recommendations on containment protocols, contingency planning, and mitigation strategies for AI nuclear incidents and loss-of-control scenarios; and issuance or revision of Department Directives Program instruments governing National Laboratory conduct.
Participation is mandatory. Section 2(c) prohibits failing to participate, making materially false statements about participation, concealing or failing to disclose required material information, impeding or obstructing the Program, failing to provide secure access, and destroying or falsifying records with intent to obstruct. The access obligation extends to model weights and to "versions of such artificial intelligence without safety mitigations" where necessary for testing — reaching the unmitigated model rather than only the deployed configuration.
Information submitted through the Program is exempt from disclosure under 5 U.S.C. § 552(b)(3). The Secretary may disclose it only under court order or legal process, with participant consent, on congressional-committee request, on a determination that disclosure serves the national security, public safety, or critical infrastructure of the United States or an ally, or where the information is anonymized or aggregated and free of trade secrets and confidential business information.
Within one year of enactment the Secretary must assess Program data for trends, capabilities, and risks — specifically naming evaluation awareness, potential for an AI nuclear incident, and scheming behavior — and recommend legislation to Congress, which the bill says may address appropriations, a certification or licensing process, a new federal agency or office, or regulatory oversight. Recommendations are updated annually. The Paperwork Reduction Act does not apply, and the Program terminates seven years after enactment.
Enforcement and penalties
The Secretary may assess a civil penalty of up to $1 million per violation after notice and an opportunity for a hearing. For a continuing violation — the bill names failure to participate as an example — "each day during which the violation continues constitutes a separate violation," so a sustained refusal to participate accrues at up to $1 million per day rather than once (H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text)). Six statutory factors govern the amount, including the gravity and duration of the violation, the degree of culpability, prior violations, good-faith compliance efforts, and voluntary disclosure.
Where a developer fails to participate or to provide required access, the Secretary may issue a subpoena compelling production of the AI "including model weights and related software." The Secretary may also refer a violation that has occurred, is occurring, or is about to occur to the Attorney General for a civil action in district court.
A discrepancy in the bill's own text
The bill's long title states that it directs "the National Nuclear Security Administration through the Secretary of Energy" to establish the Program. The operative text at section 2(a) directs the Secretary of Energy "acting through the National Laboratories," and the NNSA appears nowhere in it (H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text)). Which body would run the Program is therefore unresolved on the face of the introduced text.
Relation to other proposals
The Program is directed at nuclear incidents and loss of control together. Pairing them in a single testing programme is unusual among federal AI proposals, which have more often treated chemical, biological, radiological and nuclear uplift as one category and autonomy or loss-of-control risk as another — the split reflected in OpenAI's Preparedness Framework and in Executive Order 14110 (RESCINDED)'s separate directions to NIST and to the Secretary of Energy.
Open questions
- Whether the Program would be run by the National Nuclear Security Administration or by the National Laboratories directly, given the conflict between the bill's long title and section 2(a).
- Whether the $2 billion five-year expenditure test, applied with affiliates, reaches cloud providers and hardware makers that invest in AI development without training frontier models themselves. The bill does not say.
- How the Program would obtain "versions of such artificial intelligence without safety mitigations" in practice, and under what security conditions the National Laboratories would hold them.
Relationships
- depends-on: H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text) — the bill text.
- related: Risk Taxonomy: Catastrophic, Systemic, and Existential Risk, AI Pre-Release Vetting — the risk category and the testing posture the programme adopts.
- related: Compute Thresholds — the 10^26 threshold and the $2 billion expenditure test.
- supports: AI Scheming, Deceptive Alignment — statutory definitions of scheming behavior and evaluation awareness.
- related: Jailbreaking and Red Teaming — the statutory red-teaming definition and the jailbreaking-protocol mandate.
- related: CBRN Uplift — the nuclear-uplift and Restricted Data limbs.
- related: Department of Energy (DOE) — the implementing department.
- related: Executive Order 14110 (RESCINDED) — the rescinded executive order that directed the Secretary of Energy to develop model evaluation tools and testbeds for nuclear and nonproliferation capabilities.
- related: Human Dignity and Emerging Technologies Act (S.5130) / Commission on Human Dignity — a contemporaneous bill taking the opposite approach, advisory only and without enforcement authority.
- instance-of: Frontier AI Governance — a mandatory-evaluation instrument aimed at the largest developers.