Artificial Intelligence Risk Evaluation Act of 2025
- Sponsor
- Sen. Hawley, Josh [R-MO]
- Committees
- Commerce, Science, and Transportation Committee (primary), Homeland Security and Governmental Affairs Committee
- Last action
- Sep 30, 2026
Bottom line
This bill establishes a mandatory federal program for testing advanced AI systems for risks, with significant penalties for non-compliance, aiming to inform future comprehensive AI regulation and potentially government control.
What it actually does
This bill mandates the Secretary of Energy to establish an Advanced Artificial Intelligence Evaluation Program within 90 days to test advanced AI systems for national security, public safety, and societal risks. It requires covered AI developers to participate in this program, provide necessary materials, and prohibits them from deploying advanced AI systems without compliance. The bill also requires the Secretary to submit a plan for permanent federal oversight of advanced AI systems, including potential nationalization measures, based on insights from the program.
Proponents argue
Proponents argue that rapidly advancing artificial intelligence capabilities present significant and urgent risks to national security, public safety, economic competitiveness, and civil liberties. They contend that a secure, data-driven testing and evaluation program is essential to understand and manage these emerging risks, especially as AI approaches human-level capabilities, ensuring that future regulatory decisions are based on empirical evidence and safeguard American citizens from potential existential threats.
Opponents contend
Opponents might argue that the bill's mandatory participation, stringent requirements, and severe penalties could stifle innovation, particularly for smaller AI developers and startups, potentially driving AI research and development offshore. Concerns may also be raised about government overreach into private sector technology development, the protection of intellectual property during mandatory testing, and the broad definitions of 'adverse AI incident' and 'loss-of-control scenario' which could lead to excessive regulation.
The bill is relatively concise for its subject matter, allowing an informed reader to grasp its core provisions and implications within a reasonable timeframe.
Section 2
Sense of Congress and Purposes
This section establishes the legislative intent, stating that rapidly advancing AI presents both opportunities and significant risks to national security, public safety, economic competitiveness, civil liberties, and labor markets. It declares that the program's purpose is to provide Congress with empirical data for federal oversight and to safeguard American citizens by ensuring regulatory decisions are based on empirical testing.
Supporters argue
Supporters argue that this section clearly articulates the urgent need for government intervention in AI development, emphasizing the dual nature of AI as both an opportunity and a significant risk. They contend that establishing a clear purpose for data-driven oversight is crucial for responsible technological advancement and national security.
Critics contend
Critics might argue that while acknowledging opportunities, the 'Sense of Congress' heavily emphasizes risks, potentially setting a tone that is overly alarmist and could lead to overly restrictive regulations that stifle innovation rather than foster it.
Tradeoffs
This section balances the recognition of AI's potential benefits with a strong emphasis on its inherent risks, setting the stage for a regulatory framework that prioritizes safety and oversight over unbridled development.
Section 3
Key Definitions for Advanced AI Systems and Incidents
This section defines critical terms, including 'advanced artificial intelligence system' as one trained using more than 10^26 operations, with a mechanism for the Secretary to propose an updated definition requiring Congressional approval. It also defines 'adverse AI incident' to include loss-of-control scenarios, weaponization risks, threats to critical infrastructure, erosion of civil liberties, scheming behavior, and attempts at such incidents. 'Artificial superintelligence' is defined as AI matching or exceeding human cognitive performance and capable of self-modification to circumvent human control.
Section 4
Mandatory Participation and Penalties for Non-Compliance
This section mandates that every 'covered advanced artificial intelligence system developer' must participate in the Advanced Artificial Intelligence Evaluation Program. It requires developers to provide the Secretary with necessary materials, including underlying code, training data, model weights, and detailed information about their systems. Crucially, it prohibits any person from deploying an advanced AI system for use in interstate or foreign commerce unless they are in compliance with these participation requirements. Violations incur a penalty of not less than $1,000,000 per day.
Section 5(a)-(b)
Establishment and Activities of the Advanced AI Evaluation Program
This section requires the Secretary of Energy to establish the Advanced Artificial Intelligence Evaluation Program within 90 days. The program's activities include offering standardized and classified testing of advanced AI systems, collecting data on adverse AI incidents, implementing adversarial red-teaming, facilitating independent third-party assessments, providing formal risk reports to developers, developing containment and mitigation strategies, informing evidence-based standards, assessing the potential for artificial superintelligence, and developing proposed options for regulatory oversight, including potential nationalization.
Section 5(c)
Plan for Permanent Federal Oversight Framework
This section requires the Secretary of Energy to submit a detailed recommendation to Congress within 360 days for permanent federal oversight of advanced AI systems, based on insights and data from the evaluation program. The plan must summarize testing outcomes, recommend evidence-based standards, certification, licensing, and regulatory structures, outline proposals for continuous monitoring of AI hardware and computing resources, suggest adaptive governance strategies, and propose revisions to federal oversight (e.g., new offices or agencies). It also requires comprehensive evaluations of AI systems' potential to exceed human oversight, approach artificial superintelligence, threaten economic competition or civil liberties, and pose existential risks, including clearly articulated options for regulatory or governmental oversight measures to address imminent concerns, such as nationalization.
Allows the Secretary of Energy to propose a new definition for 'advanced artificial intelligence system' via rule, but requires a joint resolution of Congress for the new definition to take effect.
Section 3(1)(B)
Why it matters:This is standard legislative practice to provide administrative flexibility while retaining ultimate congressional oversight. It allows the definition to adapt to rapidly changing technology without requiring a full new bill each time, but ensures Congress has the final say on such a critical scope-defining term.
Case for: Proponents argue this provides necessary adaptability for the bill's core definition, allowing it to remain relevant as AI technology evolves without constant legislative amendments. The requirement for a joint resolution ensures democratic accountability.
Case against: Critics might argue that requiring a joint resolution for a technical definition change could lead to political gridlock over what should be a technical adjustment, potentially leaving the program operating with an outdated or inappropriate scope for extended periods.
Estimated impact: Potentially shifts the entire scope of regulated AI systems and developers, impacting hundreds or thousands of entities depending on future technological advancements and the new definition.
Includes in the definition of 'loss-of-control scenario' any situation where an artificial intelligence system 'otherwise behaves in an unpredictable manner so as to be harmful to humanity.'
Section 3(12)(G)
Why it matters:This phrasing is likely intended to ensure comprehensive coverage of unforeseen or novel risks that might not fit neatly into the more specific categories. Its broadness could be seen as a necessary safeguard against the unpredictable nature of advanced AI.
Case for: Proponents argue this catch-all is essential to address the inherent unpredictability of highly advanced AI systems and to cover potential harms that cannot be specifically enumerated at the time of legislation. It ensures the program can respond to emergent risks.
Case against: Critics might contend that such broad and subjective language could lead to arbitrary interpretation and enforcement, potentially allowing the government to intervene in AI development based on vague criteria, thereby creating regulatory uncertainty for developers.
Estimated impact: Significantly broadens the scope of what constitutes a 'loss-of-control scenario,' potentially increasing the types of incidents that trigger program intervention and regulatory responses.
Requires the program to develop proposed options for regulatory or governmental oversight, including 'potential nationalization or other strategic measures,' for preventing or managing the development of artificial superintelligence if it seems likely to arise.
Section 5(b)(8)
Why it matters:The concept of 'nationalization' of a private industry is extremely controversial and politically charged. Placing it within a list of program activities, rather than as a standalone section or a primary focus, may be an attempt to include it as a potential option without drawing immediate, overwhelming attention to its radical implications.
Case for: Proponents argue that given the existential risks posed by artificial superintelligence, the government must explore all possible options, including extreme measures like nationalization, to safeguard humanity. They contend that it is prudent to have such options prepared, even if they are only considered as a last resort.
Case against: Critics vehemently oppose the inclusion of 'nationalization' as a potential option, viewing it as an unprecedented and dangerous government overreach into private industry. They argue it could deter all advanced AI research and development in the U.S., violate property rights, and set a concerning precedent for government control over technology.
Estimated impact: Potentially establishes a legal and policy precedent for government seizure or control of private AI development, with profound implications for the entire technology sector and the U.S. economy.