Collaboration on Adversarial Threats and Security Risks Act
- Sponsor
- Rep. Latta, Robert E. [R-OH-5]
- Committees
- Judiciary Committee (primary)
- Last action
- Jul 23, 2026
Bottom line
This bill aims to foster collaboration among AI developers and other entities to mitigate severe AI security risks by providing a limited antitrust exemption, while retaining Attorney General oversight to prevent abuse.
What it actually does
This bill establishes a limited exemption from antitrust laws for non-Federal entities that collaborate on identifying, evaluating, or mitigating severe artificial intelligence (AI) security risks. It permits sharing information and assistance, and allows for coordinated delays or limitations on AI development or deployment, provided these actions are taken in good faith for the exclusive purpose of addressing specified AI security threats, and with prior notification to the Assistant Attorney General.
Proponents argue
Proponents argue that the rapid advancement of AI, particularly frontier models, presents novel and potentially catastrophic security risks that individual entities may not be able to address alone. This bill provides a crucial legal safe harbor for necessary collaboration to identify, prevent, and mitigate these risks, thereby safeguarding national security, critical infrastructure, and public safety without stifling innovation.
Opponents contend
Opponents might argue that this bill creates a dangerous loophole in antitrust laws, potentially allowing major AI developers to collude under the guise of security, leading to market concentration, reduced competition, and higher barriers to entry for smaller innovators. They may contend that existing antitrust frameworks are sufficient or that the exemption's scope is too broad, risking anti-competitive behavior that could harm consumers and the broader AI ecosystem.
The bill is relatively short and clearly structured, making it accessible for a reader to understand its core provisions within a reasonable timeframe.
Section 1
Short Title
This section formally names the act the 'Collaboration on Adversarial Threats and Security Risks Act.' This title succinctly communicates the bill's primary focus on enabling cooperation to address significant dangers associated with artificial intelligence.
Supporters argue
Naming the bill clearly and descriptively is standard legislative practice, ensuring easy identification and communication of its purpose.
Critics contend
No specific opposition is typically raised against this provision as it is a standard procedural element of legislation.
Tradeoffs
None inherent in the short title itself.
Section 2(6)
Defining 'Covered Artificial Intelligence Security Risk'
This section meticulously defines the specific types of threats that AI could pose, for which collaboration would be exempt from antitrust laws. These include AI being stolen or weaponized by hostile nations, substantially facilitating chemical, biological, radiological, nuclear, or offensive cyber weapons, causing critical infrastructure disruption, becoming uncontrollable, autonomously improving to create such risks, or being vulnerable to unauthorized access by hostile entities. This definition is crucial as it sets the boundaries for the antitrust exemption.
Section 3(a)(1)
Antitrust Exemption for AI Security Information Sharing
This provision states that two or more non-Federal entities can share information or assistance related to a 'covered artificial intelligence security risk' in good faith, for the exclusive purpose of addressing that risk, without violating antitrust laws. This allows companies to collaborate on identifying, evaluating, testing, analyzing, preventing, investigating, or mitigating AI threats without fear of legal repercussions under competition law, fostering a collective defense against shared risks.
Section 3(a)(2)
Antitrust Exemption for Coordinated AI Release Delays
This provision allows non-Federal entities to coordinate or enter into agreements to delay or otherwise limit the release, deployment, use, development, training, testing, or evaluation of artificial intelligence. This is permitted only if the action is for the exclusive purpose of reducing covered AI security risks and if the entities first submit written notice to the Assistant Attorney General detailing the specific risk and the scope of the proposed restriction. This acts as a 'safety brake' mechanism for the industry.
Section 3(b) and 3(c)
Conditions for Exemption and Burden of Proof
The antitrust exemption is not absolute; it only applies if the information or assistance received is used exclusively for a covered AI security purpose, and entities must implement reasonable internal controls to limit its use for other purposes. Furthermore, in any legal challenge, the non-Federal entity claiming the exemption bears the burden of proving by a preponderance of the evidence that their actions were taken in good faith and for the exclusive purpose described in the act. These clauses are critical safeguards against misuse.
Section 4
Attorney General's Authority for Injunctive Relief
This section explicitly empowers the Attorney General to seek an injunction in a court of competent jurisdiction against non-Federal entities to prohibit actions that violate antitrust laws. This power can be exercised even if an entity claims the exemption, particularly if the entity fails to meet its burden of proof, or if the Attorney General demonstrates that the actions are reasonably likely to result in an overall *increase* in covered AI security risks. This ensures a critical layer of government oversight and enforcement.
Defines 'antitrust laws' by referencing Section 102 of the Cybersecurity Information Sharing Act of 2015 (6 U.S.C. 1501) and clarifies that laws described in 6 U.S.C. 1501(2) are consistent to the extent they do not conflict.
Section 2(1)
Why it matters:It is standard legislative practice to define terms by reference to existing statutes to maintain consistency across federal law and avoid redundancy. This approach ensures that the legal scope of 'antitrust laws' aligns with established definitions in related cybersecurity legislation.
Case for: Proponents argue that defining antitrust laws by reference to existing statutes ensures legal consistency and streamlines interpretation, leveraging established legal frameworks for similar information-sharing exemptions.
Case against: Critics might contend that relying on external statutory references makes the bill less self-contained and transparent, requiring external research for a full understanding of the legal implications, potentially obscuring the full scope of the exemption for a casual reader.
Estimated impact: Clarifies the precise legal scope of the antitrust exemption, affecting all non-Federal entities operating under the bill's provisions and the legal practitioners advising them.
Defines 'covered nation' by referencing section 4872(f)(2) of title 10, United States Code, in the context of AI security risks related to hostile nations or entities owned, controlled, or directed by them.
Section 2(6)(A) and (F)(ii)(I)
Why it matters:It is standard legislative practice to define sensitive national security terms by reference to existing statutes, ensuring consistency with established defense and intelligence frameworks. This avoids redundant definitions and leverages existing legal consensus on which nations pose specific threats.
Case for: Proponents argue that referencing an existing definition for 'covered nation' ensures alignment with established national security policy, providing clarity and consistency for enforcement against foreign adversaries.
Case against: Critics might argue that requiring external research to identify these critical nations makes the bill less transparent, potentially obscuring the geopolitical scope of the AI security risks being addressed for a general audience.
Estimated impact: Determines which state actors' AI activities trigger the national security concerns addressed by the bill, thereby shaping the focus of collaborative security efforts and enforcement actions.
Any information submitted to the Assistant Attorney General pursuant to the coordinated delay/limitation provision (Section 3(a)(2)), including the written notice and any derived information, shall be used solely for Section 4 purposes, deemed voluntarily shared, exempt from disclosure under 5 U.S.C. 552 (FOIA), and must be withheld, without discretion, from the public.
Section 3(e)
Why it matters:Proponents would argue this provision is necessary to protect sensitive corporate proprietary information and national security details related to AI risks. By guaranteeing confidentiality, it encourages companies to report potential risks and coordinate actions without fear of their intellectual property or threat intelligence being disclosed to competitors or the public, which could undermine national security or commercial interests. Critics might argue it reduces transparency.
Case for: Proponents argue that confidentiality is essential to encourage companies to report sensitive AI security risks and coordinate actions without fear of proprietary information being disclosed, which could undermine national security or commercial interests and disincentivize participation.
Case against: Critics might argue that this provision significantly reduces transparency and public oversight of potentially anti-competitive agreements, making it harder for the public, media, or smaller competitors to scrutinize the justifications for coordinated delays or limitations in AI development, potentially leading to abuse.
Estimated impact: Significantly impacts the transparency of industry-government interactions regarding AI risk mitigation, potentially fostering greater industry cooperation at the cost of public disclosure.