Artificial Intelligence Safety Collaboration and Antitrust Law

Source: Congressional Research Service; CRS Legal Sidebar LSB11485 · 2026-09-28

On September 12, 2026, Anthropic chief executive Dario Amodei published a blog post arguing that the artificial-intelligence (AI) industry should slow the pace at which it develops frontier models to mitigate risks of catastrophic harm, including the loss of control of AI systems and misuse of AI for cyberattacks and bioterrorism. Amodei also proposed several measures to "pace the frontier," including coordination among AI companies to establish common safety standards and "limits on the rate of unchecked AI progress." He added, however, that "[s]ome forms of coordination that would be impactful for pacing are legally challenging, and will require government support," potentially including "waivers of antitrust restrictions."

Amodei's post—which was published amid heightened public discussion of potential AI safety risks—generated considerable commentary. Leaders of other AI labs endorsed his call to "pace the frontier." House Minority Leader Hakeem Jeffries called on Congress to take "decisive action" to slow AI development in the interest of safety. Other observers criticized Amodei's proposals. Some contended that antitrust law permits AI firms to engage in many forms of safety collaboration, making antitrust waivers unnecessary and potentially harmful. Amodei's post also generated antitrust litigation: On September 18, 2026, customers of several leading AI companies filed a putative class action alleging that the companies violated Section 1 of the Sherman Act by agreeing to slow the pace of AI innovation.

This Legal Sidebar discusses antitrust issues raised by the prospect of AI safety collaboration. It begins with an overview of the antitrust principles that govern competitor collaboration. Next, it applies those principles to evaluate the legal risks that may accompany certain types of safety collaboration between competing AI developers. The Sidebar concludes with considerations for Congress.

As the above discussion suggests, the legal standard that applies to competitor collaboration varies based on the details of the collaboration. The federal antitrust agencies—the Department of Justice (DOJ) and Federal Trade Commission (FTC)—have traditionally taken the view that joint ventures (JVs) involving research and development (R&D) typically are subject to the rule of reason. In their 2000 Antitrust Guidelines for Collaborations Among Competitors (the 2000 Guidelines), the agencies adopted that position, explaining that most competitor collaborations involving R&D are procompetitive and that such collaborations may enable participants to more efficiently develop or improve goods, services, or production processes.

Congress has also taken steps to mitigate the antitrust risks raised by R&D JVs. The National Cooperative Research and Production Act of 1993 (NCRPA) provides that certain R&D-related JVs are not per se illegal and are instead to be evaluated based on their "reasonableness, taking into account all relevant factors affecting competition." Under the NCRPA, participants in covered JVs may further limit their antitrust exposure by filing a notification disclosing certain information about a JV with the DOJ and FTC. Parties that file notifications in conformity with the NCRPA limit their antitrust exposure under federal and state law to actual damages plus costs and attorneys' fees, as opposed to the treble damages typically available under the Sherman Act.

Information sharing regarding cybersecurity enjoys special statutory protection; that protection is set to expire on December 11, 2026, absent legislative extension. The Cybersecurity Information Sharing Act of 2015 (CISA) includes an antitrust exemption allowing private entities to exchange or provide cyber threat indicators or defensive measures or assistance relating to the prevention, investigation, or mitigation of a cybersecurity threat. CISA excludes from this protection agreements to fix prices, allocate markets between competitors, monopolize or attempt to monopolize, boycott, or exchange price or cost information, customer lists, or information regarding future competitive planning.

The antitrust agencies have also issued guidance regarding the sharing of cybersecurity information. In 2014, the DOJ and FTC released a policy statement explaining that antitrust is not—nor should it be—a "roadblock to legitimate cybersecurity information sharing." While cautioning that their analysis would be fact-driven, the agencies indicated that the sharing of cyber threat information appeared "unlikely in the abstract to increase the ability or incentive of participants to raise price or reduce output, quality, service, or innovation."

Different types of safety collaboration among AI developers would raise different levels of antitrust risk under current law. The sharing of information regarding cybersecurity incidents and threats—standing alone—would likely involve low levels of antitrust risk. CISA currently offers an antitrust exemption for this type of information sharing, provided participants do not exchange specified categories of competitively sensitive information. While that exemption is slated to expire later this year absent legislative extension, courts have held that information exchanges are generally analyzed under the rule of reason. An arrangement in which AI developers share cybersecurity information may also constitute a JV entitled to rule-of-reason scrutiny under the NCRPA. While rule-of-reason review does not amount to an antitrust exemption, the sharing of cybersecurity information appears unlikely to raise significant concerns of anticompetitive harm, provided it does not involve the exchange of competitively sensitive information such as prices, costs, or production plans. The antitrust agencies' 2014 policy statement corroborates this point and suggests that enforcers are unlikely to view such conduct with suspicion.