Antitrust lawyers are raising concerns that the AI industry's push for safety cooperation among leading developers could run afoul of competition law. While companies share model evaluations, testing protocols, and access standards, and regulators encourage these efforts, the same coordination that promotes safety can also determine who gets to compete.
The tension is not new. Technology industries have long relied on coordination to set common protocols and interoperability standards. The internet itself depends on extensive cooperation among competitors. Antitrust law has recognized that such collaboration can be procompetitive. In Broadcast Music, Inc. v. CBS, the U.S. Supreme Court held that a joint licensing arrangement among competitors should be evaluated under the rule of reason, not condemned as a per se violation.
But the Court has also warned against assuming that competition is unreasonable. In National Society of Professional Engineers v. United States, it rejected an engineering society's ethical canon that treated competitive bidding as a safety threat. And in FTC v. Indiana Federation of Dentists, the Court struck down a concerted refusal to share X-rays with insurers, even though the dentists claimed they were protecting treatment quality, because antitrust law generally favors more information.
Standard-setting organizations occupy an uneasy place in this legal framework. The same standards that enable interoperability can also pick winners and losers. In American Society of Mechanical Engineers, Inc. v. Hydrolevel Corp. and Allied Tube & Conduit Corp. v. Indian Head, Inc., the Supreme Court confronted cases where participants allegedly manipulated a standard-setting process to disadvantage a rival technology. The lesson for AI is that safety standards can be captured by firms with a competitive stake in the outcome.
Safety and Competition
AI safety frameworks bring the antitrust question into sharp focus because the topics most likely to attract industry coordination-model evaluation requirements, capability thresholds, access restrictions-also directly shape market structure. If leading developers agree on common testing protocols before deploying frontier models, the arrangement could improve transparency and regulatory oversight. But compliance costs rarely fall evenly. Testing frameworks manageable for large incumbents may prove much more burdensome for startups.
A capability-threshold agreement, where developers pledge not to release models above a certain level until they meet safety requirements, would also marginalize new entrants. Evaluation and monitoring regimes built around systems served through an API would extinguish open-source developers who distribute model weights directly. Decisions about what constitutes sufficient testing or acceptable risk give the firms at the table a say over which competitors get to release a frontier model and when.
Likely Future Disputes
The analysis grows harder when safety cooperation extends to collective access decisions-when a group of leading developers determines that certain models should not be released publicly or that access should be limited to particular users. Testing requirements affect the cost and pace of competition. Access restrictions can determine whether competition occurs at all.
Membership and certification rules in other industries have raised the same issue. In Realcomp II Ltd. v. FTC, a multiple listing service's rules on displaying property listings were found to disadvantage discount brokerages, even though the rules were framed in neutral, operational terms. Antitrust law's response has focused on whether the resulting restriction is reasonably necessary to its stated purpose and applied without favoring incumbents over new entrants.
The difficult cases won't look like price-fixing. They'll look like a safety framework that, on its own terms, is hard to argue against-but happens to leave the firms that designed it better positioned than the ones that didn't.
Why this matters for legal professionals
As policymakers continue to encourage cooperation on AI safety, antitrust lawyers will increasingly be asked when coordination that benefits the industry becomes coordination that limits competition. For now, no one has challenged an AI safety arrangement on these grounds. The first case to do so will set the template for everything that follows.
Legal professionals advising AI developers or regulators need to scrutinize whether safety collaborations remain reasonably tied to legitimate objectives and whether the standards are structured to serve competitive interests. Resources such as AI for Legal Professionals Courses can help practitioners stay current on these emerging intersections of technology and competition law.
Your membership also unlocks:
