Antitrust safe harbor for AI security cooperation

Full Title:
Collaboration on Adversarial Threats and Security Risks Act

Summary#

This bill creates a limited antitrust safe harbor for private and other non‑federal groups that share information or coordinate actions to address serious security risks from advanced artificial intelligence models. The main change is that certain cooperative actions that would otherwise raise antitrust questions are not treated as antitrust violations if they are done in good faith and only to reduce covered AI security risks. The bill aims to encourage industry cooperation to prevent theft, misuse, or dangerous deployment of powerful AI.

  • Allows two or more non‑Federal entities to share information or assistance about covered AI security risks without being treated as breaking antitrust laws, if done in good faith and for that exclusive security purpose.
  • Allows coordination to delay or limit release, deployment, testing, training, or evaluation of AI to reduce security risks, but requires written notice to the Assistant Attorney General in advance.
  • Requires recipients to use shared information only for the security purpose and to have reasonable internal controls to prevent other uses.
  • Makes the antitrust safe harbor an affirmative defense (the entity claiming it must prove good faith and exclusive purpose by a preponderance of the evidence).
  • Exempts notices to the Assistant Attorney General from public disclosure (FOIA) and limits how that submitted information can be used.
  • Explicitly does not permit traditional antitrust violations like price‑fixing, market allocation, monopolization, or sharing price information.

What it means for you#

  • AI companies, developers, and model owners: This could make it safer under antitrust law to share vulnerability information or to jointly delay a model release, provided you do so only to reduce specified security risks and follow the notice and control rules.
  • Industry groups, consortia, researchers, and labs: You would be able to coordinate certain security measures with peers without automatically triggering antitrust liability, but you must document good faith purpose and keep internal limits on use of shared material.
  • Lawyers and compliance officers for private entities: You would need to prepare to carry the burden of proving the exclusive security purpose if challenged, set up internal controls, and submit required written notices to the Antitrust Division before certain coordinated delays or limits.
  • Federal government / Department of Justice (Antitrust Division): The Assistant Attorney General would receive advance notices of coordinated delays or limits and could seek injunctions if the DOJ concludes the conduct is not justified or increases risks.
  • General public / security officials: The bill is intended to increase private cooperation to reduce risks from misuse, theft, or dangerous deployment of AI models that could harm national security or public safety.

Expenses#

No publicly available information.

  • The bill may increase administrative workload for the Antitrust Division because it requires the Assistant Attorney General to receive and handle written notices and to review whether coordinated actions are acceptable.
  • Private entities may face compliance costs to create internal controls, document good faith security purposes, and prepare notices.
  • No formal budget estimate or fiscal note is included in the available material.

Proponents' View#

  • The bill appears intended to encourage companies and other non‑federal actors to share information and cooperate to reduce serious AI security risks without fearing automatic antitrust liability.
  • Supporters may argue this could improve detection of theft, vulnerabilities, and misuse of frontier AI, and make it easier to coordinate protective measures such as delaying risky releases.
  • The bill includes safeguards (notice to DOJ, requirement of exclusive security purpose, internal controls, and FOIA exemption for notices) that supporters could view as balancing collaboration with oversight and confidentiality.
  • The bill targets severe risks spelled out in the text, such as model theft by hostile states, facilitation of weapons development, harm to critical infrastructure, and loss of human oversight.

Opponents' View#

  • One concern is that the bill does not give a detailed test for “exclusive purpose,” so it may be unclear when cooperation is truly limited to security and when it covers other competitive aims.
  • The requirement that entities prove an affirmative defense (by a preponderance of the evidence) could leave companies exposed to costly litigation before they can rely on the safe harbor.
  • The exemption from public disclosure for notices could reduce transparency about why companies delay or limit product releases.
  • The bill does not fully explain how the Antitrust Division will assess notices or how quickly it must act, which may create uncertainty or delay for companies trying to follow the rules.
  • There is a risk that coordination described as security‑related could be used to mask anti‑competitive behavior unless DOJ oversight is active and effective.