Artificial Intelligence Risk Management and Security Act of 2026
- Sponsor
- Sen. Warner, Mark R. [D-VA]
- Committees
- Commerce, Science, and Transportation Committee (primary)
- Last action
- Sep 29, 2026
Bottom line
The bill creates a new regulatory framework to proactively manage the safety and security risks posed by advanced artificial intelligence systems, with mandatory compliance and enforcement mechanisms.
What it actually does
This bill establishes an Artificial Intelligence Safety Board within the Department of Commerce to address AI risks, particularly for 'frontier artificial intelligence models.' It mandates that developers comply with technical standards set by the Board, develop and submit safety plans, and report AI security and safety incidents. The bill also directs the National Institute of Standards and Technology (NIST) to create a publicly accessible database of AI incidents and develop profiles and templates for agentic AI systems.
Proponents argue
Proponents argue that this bill is a crucial step to proactively address the rapidly evolving risks associated with advanced artificial intelligence, protecting national security, public health, and safety. They contend that establishing clear standards and a dedicated oversight body will foster responsible innovation and prevent catastrophic failures, ensuring the United States remains a leader in safe AI development.
Opponents contend
Opponents might contend that the bill imposes significant regulatory burdens on AI developers, potentially stifling innovation and competitiveness, especially for smaller companies. Concerns could also be raised about the government's mandatory access to proprietary model information, such as weights and configuration files, which could pose intellectual property risks and create a chilling effect on development.
The bill is relatively concise for its scope, allowing an informed reader to grasp its core provisions within a reasonable timeframe, though the technical nature of AI risk management requires careful attention.
Section 3(a)-(e)
Establishment and Duties of the Artificial Intelligence Safety Board
This provision establishes the Artificial Intelligence Safety Board within the Department of Commerce as a permanent advisory committee. The Board is tasked with developing technical evaluations for high-risk AI capabilities, creating processes and metrics for evaluating risks from frontier AI models, and establishing technical standards for model cards, cybersecurity, personnel risk, and secure testing environments. Developers of frontier AI models are required to comply with these standards, with civil penalties for violations.
Supporters argue
Supporters argue that a dedicated, expert-led board is essential to keep pace with the rapid advancements in AI and to proactively develop necessary safety and security standards. They contend that mandatory compliance is critical to ensure that all developers meet a baseline level of risk management, protecting the public and national interests.
Critics contend
Critics may argue that establishing a new board with broad standard-setting authority could lead to regulatory overreach and stifle innovation. They might express concern that the compliance burden and potential penalties could disproportionately affect smaller developers or create barriers to entry in the AI market.
Tradeoffs
This provision balances the need for robust AI safety and security oversight with concerns about potential regulatory burdens on the rapidly evolving AI industry. It also navigates the tension between government authority in setting technical standards and industry's desire for flexibility and self-regulation.
Section 3(f)
Mandatory Pre-Release Access for Frontier AI Models
This provision requires developers of frontier artificial intelligence models to provide the Artificial Intelligence Safety Board with access to their models, including weights, configuration files, runtimes, or software libraries, not later than 45 calendar days before introducing the model into interstate or foreign commerce. This allows the Board to conduct its own evaluations prior to public release.
Section 4
Model Safety Plan Requirement
This section mandates that each developer create, publish, and adhere to a 'Model Safety Plan' for every artificial intelligence system or model they create, substantially modify, or use in training other models. Each plan must include an AI capabilities or risk assessment, specific mitigation measures for identified risks across the product lifecycle, and identify the corporate officer responsible for its implementation. These plans must be submitted to the Secretary of Commerce.
Section 5
Database for Artificial Intelligence Security and Safety Incidents, Flaws, and Risks
This section directs NIST to establish mechanisms for voluntary and mandatory reporting of AI security and safety incidents and flaws. It mandates the creation of a publicly accessible database of these incidents and flaws, ensuring anonymity for affected parties unless consent is given. Frontier AI model developers/providers and critical infrastructure operators using AI are required to report incidents within 30 days, or 72 hours if there's an imminent threat to national security, critical infrastructure, or public safety.
Section 6
Agentic Artificial Intelligence Trust and Verification
This section directs NIST to develop an 'Agentic AI Profile' as a companion to its AI Risk Management Framework, specifically for artificial intelligence agents (AI systems capable of executing actions directly). This profile will define novel risks, classify agent autonomy levels, and address cybersecurity risks unique to AI agents. NIST is also tasked with developing a common template for documenting AI agents and their evaluations against recognized standards, aiming for cross-sector applicability.
The Secretary of Commerce, in coordination with the Secretary of State, is directed to undertake negotiations with foreign partners and enter into agreements to facilitate cooperative activities, regulatory reciprocity, and intellectual property protection related to AI safety and security risks. This includes establishing joint testing environments, conducting joint research programs, promulgating joint advisories, facilitating secure information sharing, and standardizing evaluation protocols.
Section 3(a)(2)(C)
Why it matters:This is standard legislative practice for detailing specific functions or authorities related to a broader organizational structure. It's not necessarily evasive but could be overlooked as a significant policy directive.
Case for: Proponents argue that international cooperation is essential for effective AI governance, as AI risks are global and cannot be contained by national borders. Joint efforts can leverage diverse expertise, harmonize standards, and prevent regulatory arbitrage, fostering a safer global AI ecosystem.
Case against: Critics might argue that entering into international agreements, especially on sensitive topics like IP protection and regulatory reciprocity, can be complex, time-consuming, and potentially compromise national interests or sovereignty if not carefully managed. There could be concerns about sharing sensitive information with foreign entities.
Estimated impact: Potentially significant impact on global AI governance and international relations, shaping how AI safety and security are managed worldwide.
This clause mandates that the Secretary of Commerce shall sponsor each member of the Artificial Intelligence Safety Board for a security clearance at the Top Secret level with access to sensitive compartmented information (SCI). Any member who fails to obtain this clearance, or is denied access to classified information, shall be removed from the Board and replaced.
Section 3(a)(4)(A)(iii)
Why it matters:This is a technical requirement for handling classified information, which is standard for bodies dealing with national security matters. Its placement reflects its procedural nature rather than an attempt to hide it.
Case for: Proponents argue that given the national security implications of advanced AI and the Board's duties, access to Top Secret/SCI information is absolutely necessary for members to make informed decisions and develop effective standards. This ensures the Board can address the most critical risks.
Case against: Critics might argue that such a high security clearance requirement could severely limit the pool of eligible experts, potentially excluding valuable technical talent or individuals from civil society who may not be able or willing to obtain such clearances. This could lead to a less diverse or less representative Board.
Estimated impact: Significant impact on the composition and expertise of the Artificial Intelligence Safety Board, potentially narrowing the range of individuals who can serve.
Information shared voluntarily or mandatorily with NIST regarding AI security or safety incidents and flaws shall be exempt from public disclosure under FOIA (5 U.S.C. 552(b)(3)(B)) and other laws, and shall not be deemed a waiver of any applicable privilege or protection, including trade secret protection, unless an affected party consents to inclusion of their names in the public database.
Section 5(a)(5)(B)
Why it matters:This is a standard legislative technique for providing specific legal protections for information sharing, often placed within the section that mandates or encourages such sharing. It's intended to facilitate compliance rather than obscure the provision.
Case for: Proponents argue that this exemption is essential to encourage voluntary reporting of incidents and flaws, as companies would be hesitant to share sensitive information if it could be publicly disclosed or used against them. Protecting trade secrets and ensuring anonymity fosters trust and cooperation, leading to a more comprehensive understanding of AI risks.
Case against: Critics might argue that broad exemptions from public disclosure could reduce transparency and accountability, making it harder for the public or independent researchers to scrutinize AI incidents or the effectiveness of regulatory responses. They might contend that too much secrecy could undermine public trust.
Estimated impact: Significant impact on the willingness of AI developers and operators to report incidents, directly affecting the comprehensiveness and utility of the AI incident database.