Companies could claim limited legal protection for AI security cooperation

Nonfederal groups could claim protection under laws that protect competition when sharing AI security information or help. Only sharing tied to the bill’s listed AI security risks would qualify. Sharing must be honest and almost entirely for addressing listed AI security risks. Recipients must use the information or help for that purpose and maintain reasonable safeguards against other uses. This protection would not permit price-fixing, dividing markets, monopolizing, boycotts, or sharing price or cost information.

Nonfederal groups could claim protection under competition laws for agreements to delay or limit AI activities. Agreements could restrict AI activities such as development, testing, and release. The agreement must be almost entirely for reducing the bill’s listed AI security risks. Written notice must describe the specific risk and the scope of the proposed restriction. Groups must submit notice before undertaking the coordinated delay or limitation.

The Attorney General could ask a court to stop conduct that violates competition laws. These limits would apply to groups claiming either sharing protection or protection for coordinated restrictions. Groups claiming protection must show it is more likely than not that they acted honestly and almost entirely for the required security purpose. Protection would not block a court order if groups fail that test or the Attorney General shows their actions would reasonably likely increase overall listed AI security risks.

The Justice Department would withhold submissions about coordinated AI restrictions from federal public-records requests. This would cover notices, other submitted information, and information derived from submissions that reveals their substance. The department could use the information only for the bill’s court-order process. It would have no discretion to release the covered information under federal public-records law.