Sponsors (2)
The bill creates a legal framework to enable confidential coordination and stronger enforcement against unsafe or anticompetitive AI conduct—improving national security and accountability—but at the cost of increased compliance and litigation burdens, reduced transparency, and potential chill on research, innovation, and competition, especially for researchers and small firms.
Developers, companies, and government agencies can legally share AI security information, delay or limit risky AI releases after notifying DOJ, and submit vulnerability reports without public disclosure, enabling faster coordinated mitigation of AI threats and encouraging candid reporting.
AI developers and operators get clearer statutory definitions of security risks and 'unauthorized access' (aligned with existing statutes), reducing regulatory uncertainty and helping teams design compliant security controls.
The Attorney General is empowered to seek injunctions against non‑federal actors engaging in anticompetitive or unsafe AI conduct and the bill clarifies that invoking an 'AI security' purpose does not automatically block court orders, increasing the government's ability to stop harmful deployments before damage occurs.
Researchers, scientists, and small AI firms face a chill on legitimate research, collaboration, and cross‑border work because broad definitions of 'assistance' and 'unauthorized access', cross‑references to national‑security statutes, and a narrow 'exclusive purpose' test could bring routine data sharing or dual‑use research within regulatory or enforcement risk.
Model owners and smaller companies may face substantially higher compliance and legal costs — implementing stricter access controls, auditing, and incident response — and greater litigation risk (including the burden to prove exemption), raising barriers for startups and smaller teams.
The security-exemption mechanisms could be misused to justify anticompetitive coordination (including timing of product releases) while FOIA exemptions and narrow antitrust prohibitions reduce transparency and leave tacit-collusion risks hard to detect, harming competition and consumers.
Based on analysis of 4 sections of legislative text.
Allows limited antitrust-safe collaboration among non-federal entities to share information or delay AI activities solely to address defined AI security risks, with notice to DOJ for coordination.
Official title: Establish the applicability of antitrust laws to the sharing of artificial intelligence frontier model risks, and for other purposes.
Introduced July 23, 2026 by Adam Schiff · Last progress July 23, 2026
Creates a narrowly tailored antitrust safe harbor to let non-federal entities share information and coordinate briefly when the sole purpose is to prevent or mitigate defined AI security risks. It sets definitions for covered AI risks and assistance, makes the safe harbor an affirmative defense with a preponderance-of-evidence burden, requires prior notice for coordinated delays or limits, protects voluntary submissions from public disclosure, and preserves DOJ authority to seek injunctions where harms remain or the defense fails. The law applies to non-Federal entities (companies, researchers, other organizations) acting in good faith for covered AI security purposes, while explicitly not protecting classic antitrust violations like price-fixing, market allocation, or monopolization. It also defines unauthorized access to AI (including model extraction and integrity attacks) and lists specific categories of covered national-security and infrastructure risks that justify collaboration under the safe harbor.