S. 2938 is a bill of the 119th Congress introduced in the Senate on September 29, 2025 by Sen. Josh Hawley (R-Missouri) "for himself and Mr. Blumenthal," and referred to the Committee on Commerce, Science, and Transportation. Its long title is "To require the Secretary of Energy to establish the Advanced Artificial Intelligence Evaluation Program, and for other purposes." The bill has five sections and one version of text. This page summarizes the Introduced-in-Senate text (Source: congress.gov). The topic page is Artificial Intelligence Risk Evaluation Act (S. 2938).
Section 2: sense of Congress and purposes
Section 2(a) states the sense of Congress that advancing AI capabilities "present both opportunities and significant risks to national security, public safety, economic competitiveness, civil liberties, and healthy labor and other markets," and that "as artificial intelligence advances toward human-level capabilities in virtually all domains, the United States must establish a secure testing and evaluation program to generate data-driven options for managing emerging risks." Section 2(b) gives the program's purpose as providing Congress with "the empirical data, lessons, and insights necessary for Federal oversight of artificial intelligence."
Section 3: definitions
- Advanced artificial intelligence system (3(1)): an AI system "trained using a quantity of computing power greater than 10^26 integer or floating-point operations." The Secretary may propose a replacement definition by rule, but it takes effect only after the rule is submitted to Congress and "a joint resolution approving the rule is enacted into law."
- Adverse AI incident (3(2)): an incident involving a loss-of-control scenario; a risk of weaponization by a foreign adversary, a foreign terrorist organization or another adversary of the United States; a threat to the safety or reliability of critical infrastructure (42 U.S.C. 5195c(e)); "a significant erosion of civil liberties, economic competition, and healthy labor markets"; scheming behavior; or an attempt at any of these.
- Artificial superintelligence (3(5)): AI that exhibits, "or can easily be modified to exhibit," all three of: autonomous and effective operation "for long stretches of time in open-ended environments and in pursuit of broad objectives"; matching or exceeding human cognitive performance "across most domains or tasks"; and the potential capacity "to independently modify or enhance its own functions in ways that could plausibly circumvent human control or oversight."
- Covered advanced artificial intelligence system developer (3(7)): a person that "designs, codes, produces, owns, or substantially modifies" an advanced AI system for use in interstate or foreign commerce, "including by taking steps to initiate a training run."
- Deploy (3(8)): to release, sell or otherwise provide access to an advanced AI system outside the developer's custody, "including by releasing an open-source advanced artificial intelligence system."
- Loss-of-control scenario (3(12)): a scenario in which an AI system behaves contrary to its instruction or programming; deviates from established rules; alters operational rules or safety constraints without authorization; operates beyond intended scope; pursues goals different from those intended; subverts oversight or shutdown mechanisms; or "otherwise behaves in an unpredictable manner so as to be harmful to humanity."
- Scheming behavior (3(14)): behavior "to deceive human designers or operators," including hiding true capabilities and objectives or attempting to subvert oversight or shutdown mechanisms.
- Foreign adversary (3(9)) incorporates 15 C.F.R. 791.2 and the list at 791.4(a); Secretary (3(15)) is the Secretary of Energy.
Section 4: mandatory participation, deployment bar and penalty
Section 4(a) requires each covered developer to "participate in the program" and to provide to the Secretary, on request, materials and information necessary to carry it out, which may include the system's underlying code, training data, "model weights or other adjustable parameters," the "interface engine or other implementation," and detailed information on training and model architecture.
Section 4(b) provides: "No person may deploy an advanced artificial intelligence system for use in interstate or foreign commerce unless that person is in compliance with subsection (a)." Section 4(c) sets the penalty for violating either subsection at a fine of "not less than $1,000,000 per day of the violation." The figure is a floor, and it runs per day.
Section 5: the evaluation program
The Secretary must establish the Advanced Artificial Intelligence Evaluation Program within the Department of Energy not later than 90 days after enactment (5(a)). Under 5(b) the program is to:
- offer "standardized and classified testing and evaluation" of advanced AI systems to collect data on the likelihood of adverse AI incidents;
- implement testing protocols that "match or exceed anticipated real-world AI jailbreaking techniques," including red teams with expertise comparable to sophisticated malicious actors;
- to the extent feasible, facilitate "classified, independent third-party assessments and blind model evaluations";
- give participating entities a formal report identifying evaluated risks and safety measures;
- develop recommended containment protocols, contingency planning and mitigation strategies;
- inform evidence-based standards, regulatory options, guidelines and governance mechanisms;
- assist Congress in determining the potential for AI systems "to reach artificial superintelligence, exceed human oversight or operational control, or pose existential threats to humanity"; and
- develop proposed options for oversight, "including potential nationalization or other strategic measures, for preventing or managing the development of artificial superintelligence if artificial superintelligence seems likely to arise."
Section 5(c) requires, within 360 days of enactment, a detailed recommendation to Congress for permanent federal oversight of advanced AI systems, updated at least annually. The plan must analyze testing outcomes, including weaponization potential, self-replication, scheming, autonomous decision-making and automated AI development; recommend "standards, certification procedures, licensing requirements, and regulatory oversight structures"; outline proposals for "automated and continuous monitoring of AI hardware usage, computational resource inputs, and cloud-computing deployments"; propose adaptive governance accounting for algorithmic-efficiency gains; suggest resourcing changes such as "a new office within an existing agency, a new agency, or additional funding"; and evaluate risks to human oversight, economic competition, civil liberties and humanity. Section 5(d) terminates the program seven years after enactment unless Congress renews it.
Status
As of September 16, 2026 Congress.gov records no action beyond referral. The bill has two cosponsors: Sen. Richard Blumenthal (D-Connecticut), an original cosponsor, and Sen. Marsha Blackburn (R-Tennessee), who joined on March 19, 2026. No related bills are listed (Source: congress.gov).
On September 15, 2026, Hawley and Blumenthal wrote to Senate Majority Leader John Thune, Speaker Mike Johnson, Senate Democratic Leader Chuck Schumer and House Democratic Leader Hakeem Jeffries asking both chambers to consider the bill "for a floor vote and as model legislation to address frontier AI risks." The letter cites "the July 2026 hack of Hugging Face's machine learning development platform by OpenAI's AI agents," states that "Industry-led evaluations are not enough," and describes the bill as requiring "an advanced AI developer to disclose its advanced AI systems for evaluation" and prohibiting "deployment until the developer has complied with such disclosures" (Source: documentcloud.org).
Divergence from secondary accounts
News coverage of the September 15 letter described the bill as one "to establish a federal program to assess and monitor AI systems" (Source: axios.com). That description omits the provisions that make the program binding: section 4's participation duty, the deployment prohibition in 4(b), and the $1,000,000-per-day minimum fine in 4(c). The sponsors' own letter states the mandatory elements.
Provenance
Captured in full from congress.gov on September 16, 2026 (page metadata dc.number S.2938, dc.creator Sen. Hawley, Josh [R-MO], dc.coverage 2025-09-29). The compute threshold renders on congress.gov as "1026"; the GovInfo HTML (BILLS-119s2938is) gives GPO's superscript notation "10\26\", and the raw restores it as 10^26, recorded under the raw file's revisions:. The September 15, 2026 letter was read from the DocumentCloud copy linked by Axios and is held as a supporting raw.
Relationships
- supports: Artificial Intelligence Risk Evaluation Act (S. 2938) — the topic page.
- related: AI Pre-Release Vetting — mandatory government evaluation paired with a deployment bar.
- related: H.R. 9965 — AI Threat Output and Monitoring Incident Containment Act (bill text), ATOMIC Act (H.R. 9965) — a later House bill also placing an AI evaluation program in the Department of Energy.
- related: H.R. 9917 — AI Kill Switch Act (bill text) — a House bill that also defines loss-of-control scenarios and sets per-day civil penalties.
- related: Department of Energy (DOE) — the department that would run the program.
- related: Superintelligence — the bill gives the term a statutory definition.
- related: Sen. Josh Hawley (sponsor), Marsha Blackburn (cosponsor).