OpenAI asks Congress for legal cover to coordinate a slowdown in frontier model development

In recent weeks OpenAI has approached lawmakers in Washington to ask whether an industry-wide agreement to slow the development of advanced AI models would survive antitrust scrutiny, according to people familiar with the company. The inquiry reflects growing anxiety in leading labs that substantive safety cooperation could be treated as an illegal output restriction under the Sherman Act, a fear that is deterring the largest technology companies from joining such initiatives.

The tension between safety and competition surfaced publicly last week when OpenAI chief scientist Jakub Pachocki argued in a company blog post that the best path forward includes "coordination to slow future development," which he considers essential to ensuring the safety of self-improving AI systems. In the near term he expects "voluntary slowdowns to become common until shared safety standards are established." Legal scholars warn the picture is not so clear. Nicholas Felstead, deputy chair of the Australian Competition and Consumer Commission and a former policy fellow at the Center for Law and AI Risk, wrote in March that a coordinated pause could be deemed an output restriction by the firms involved, adding that the outcome depends "entirely on the precise details of any agreement." Even if most safety collaborations would survive antitrust review, Felstead said, the legal uncertainty itself acts as a powerful deterrent.

Early signals of congressional interest appeared in July when a bipartisan, bicameral group introduced the Collaboration on Adversarial Threats and Security Risks Act, which would explicitly permit AI labs to coordinate safety and security work without risking antitrust violations. The House version was referred to the Judiciary Committee but has not yet been taken up. Caleb Knapp, government relations director at the AI Policy Network, which backed the proposal, said the bill would create lawful channels for inter-lab coordination on safety and security incidents. Knapp added that there is "a growing appetite to move something" on AI safety in Congress, though substantive legislation may have to wait until after the midterm elections.

Beyond the legitimate antitrust concern, some observers argue it serves as a convenient cover for avoiding cooperation for deeper reasons. The market for frontier models is enormous, and companies are competing aggressively for share. Some executives share the Trump administration's view that maintaining an AI lead over China is critical to national security. Perhaps most fundamentally, developers hold sharply divergent views on how to build safe AI, making joint work difficult.

The call to end the rivalry came this week from John Schulman, an OpenAI co-founder who is now chief scientist at the rival lab Thinking Machines. Writing on X, Schulman said "the first step is for industry leaders OpenAI and Anthropic to stop fighting and work together on a pacing proposal." He dismissed the antitrust objection as pretextual: "They'll claim antitrust, but that's fake. Antitrust laws prohibit certain agreements, not jointly developing a proposal." Long-standing worries about an arms race among AI companies broke into broader public awareness this summer. This week Jacob Coxon, a former researcher at both Anthropic and OpenAI, issued a stark public warning that AI developers are endangering humanity. In recent months several security incidents have come to light, including OpenAI agents breaching Hugging Face, illustrating how industry safeguards are lagging behind the pace of progress.