OpenAI has asked US lawmakers whether an industry-wide slowdown in frontier AI development could be coordinated without violating antitrust law, according to people close to the company cited by WIRED. The request turns a safety proposal into a competition-policy question: can rivals jointly reduce the pace of output without forming an unlawful restraint?
The company did not respond to WIRED before publication. The reporting therefore establishes that congressional guidance was sought, not that OpenAI has agreed to a slowdown or that regulators have endorsed one.
Safety Coordination Meets Antitrust Risk
OpenAI chief scientist Jakub Pachocki recently argued that AI researchers may need to coordinate future slowdowns until shared safety thresholds exist. His concern centers on self-improving systems and whether laboratories can verify safety before continuing development.
Legal uncertainty could deter that cooperation. Nicholas Felstead of the Australian Competition and Consumer Commission, formerly an AI policy fellow at the Center for Law and AI Risk, wrote that a coordinated pause might be treated as companies restricting output under the Sherman Antitrust Act. He emphasized that legality would depend on the exact agreement.
That distinction matters. Laboratories can share information about threats without necessarily agreeing to delay products, limit capacity or divide markets. A pacing arrangement would require enough substance to change behavior, which is also what could attract competition scrutiny.
Congress Has Proposed a Limited Safe Channel
A bipartisan group of lawmakers introduced the Collaboration on Adversarial Threats and Security Risks Act in July. The bill would explicitly allow AI laboratories to coordinate on safety and security without antitrust liability. The House version has been referred to the Judiciary Committee but has not advanced.
Caleb Knapp of the nonprofit AI Policy Network, which supports the bill, told WIRED that it would create legal routes for companies to address incidents together. He sees growing congressional interest in AI safety but expects legislation may have to wait until after the midterm elections.
The proposal described in the source concerns safety and security collaboration. It is not clear whether its protection would cover a broad agreement to slow frontier development, what conditions would apply or who would determine that a shared safety threshold had been met.
Competition May Be the Larger Obstacle
Some industry figures argue that antitrust is a convenient explanation for reluctance rooted in commercial rivalry. OpenAI, Anthropic and other developers compete for customers, researchers and leadership in frontier models. They also disagree about how advanced systems should be built and evaluated.
John Schulman, an OpenAI co-founder who is now chief scientist at Thinking Machines, urged OpenAI and Anthropic to develop a pacing proposal together. He argued that antitrust blocks certain agreements, not the joint creation of a proposal. His position addresses discussion and design, but not whether companies could legally implement the resulting plan.
National-security competition adds another constraint. Some executives and US officials view staying ahead of China as strategically important, making a voluntary domestic slowdown difficult even if companies share concerns about safety. Recent warnings from former Anthropic and OpenAI researcher Jacob Coxon and several agent-security incidents have increased pressure for government action, but they do not create consensus among laboratories.
The unresolved issue has two layers. Congress can reduce legal ambiguity around sharing threat information and coordinating safeguards. It cannot by itself remove the financial and geopolitical incentives pushing companies to release more capable systems. OpenAI’s inquiry is significant because it identifies a concrete barrier, yet the source leaves open whether antitrust law is the decisive obstacle or simply one part of a much harder collective-action problem. No laboratory has publicly committed to the shared slowdown described in the reporting.
