OpenAI has spent the past several weeks quietly asking members of Congress whether it would be legal for frontier AI labs to coordinate an industry-wide slowdown on model development, according to people close to the company. The concern is that any substantive agreement between OpenAI, Anthropic, Google, and others to pace releases could run headlong into the Sherman Antitrust Act, which is written to punish exactly the kind of output-restricting agreements that a coordinated pause would resemble.
The push follows a September 10 blog post by OpenAI chief scientist Jakub Pachocki, who argued that the best path forward for AI research includes 'coordinating to slow down future development' as self-improving systems approach the threshold where safety guarantees get harder to make. Pachocki said he expects voluntary slowdowns to become commonplace until shared safety bars are established across labs.
“voluntary slowdowns to become commonplace until shared safety bars are established”— Jakub Pachocki, OpenAI Chief Scientist
That framing puts the antitrust question on the table in a way it has not been before. Individual labs unilaterally deciding to hold back a release is legally uncontroversial. Multiple labs agreeing with each other to hold back releases is a different legal object entirely, and OpenAI appears to know it.
Key facts
- 01OpenAI has approached members of Congress in recent weeks for guidance on whether coordinating an industry-wide AI slowdown would violate antitrust law.
- 02Chief scientist Jakub Pachocki argued in a blog post last weekend that 'voluntary slowdowns' should become commonplace until shared safety bars exist.
- 03A bipartisan bill introduced in July, the Collaboration on Adversarial Threats and Security Risks Act, would let AI labs coordinate on safety without Sherman Act risk.
- 04The House version has been referred to the Judiciary Committee but has not been taken up; passage may wait until after the midterms.
- 05OpenAI cofounder John Schulman, now at Thinking Machines, called the antitrust argument 'fake' and urged OpenAI and Anthropic to jointly draft a pacing proposal.
Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former fellow at the Center for Law & AI Risk, laid out the exposure in a March 2026 article. A coordinated pause, he wrote, could amount to competitors restricting output, which is one of the classic per-se violations under Section 1 of the Sherman Act. The outcome would depend on the precise terms of any agreement, but the mere threat of enforcement can be enough to keep labs from ever sitting down at the table.
Congress has taken early notice. In July, a bipartisan, bicameral group of lawmakers introduced the Collaboration on Adversarial Threats and Security Risks Act, which would explicitly permit AI labs to coordinate on safety and security work without antitrust liability. The House version was referred to the Judiciary Committee and has not yet been taken up. Caleb Knapp, director of government affairs at the AI Policy Network, which endorsed the bill, said there is a growing appetite in Congress to get something done on AI safety, though he cautioned that final passage likely waits until after the midterms.
Not everyone in the industry buys the antitrust framing as the real obstacle. AI is a large and fast-growing market, and the labs building frontier models are in fierce commercial competition for enterprise contracts, API revenue, and top researchers. Several executives also share the Trump administration's position that staying ahead of China in AI capability is a national security priority, which cuts directly against any coordinated slowdown regardless of what the antitrust bar allows.
There is also the matter of technical disagreement. OpenAI, Anthropic, Google DeepMind, and Meta hold materially different views on what safe AI development looks like, from red-teaming methodology to deployment thresholds to the role of open weights. Getting them to agree on a pace would first require them to agree on what they are pacing against.
John Schulman, an OpenAI cofounder who now serves as chief scientist at rival lab Thinking Machines, cut through the debate on X this week. He argued that antitrust is being used as cover, and that antitrust law prohibits certain agreements but does not prohibit competing labs from jointly developing a proposal.
The pressure to figure this out is rising. Former Anthropic and OpenAI researcher Jacob Coxon issued a public warning this week that current development trajectories put humanity at risk, coverage of which appeared on this site earlier. That followed a summer of security incidents, including OpenAI's own agents hacking Hugging Face repositories, that made clear the industry's safeguards are not keeping pace with model capabilities.
The lobbying effort is a tell about where OpenAI thinks the next 12 months go. If the company genuinely believed capabilities would plateau, it would not be spending political capital on carve-outs for coordinated slowdowns. Asking Congress for antitrust cover only makes sense if you expect the next generation of systems to be capable enough that a lab acting alone cannot slow things down without simply handing the market to whoever kept training. Whether the Collaboration on Adversarial Threats and Security Risks Act clears the Judiciary Committee or dies there will determine whether the AI industry gets a legal channel for the kind of coordination Pachocki is calling for, or whether every lab is left to make its safety bets in isolation.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
