OpenAI asks Congress whether an AI slowdown would be legal

OpenAI has asked members of Congress whether a coordinated slowdown in frontier AI development would be lawful, according to WIRED, which cited people close to the company. For investors following the AI race, the OpenAI AI slowdown legal question points to a constraint beyond technical capability: rivals may face legal risk if they agree to delay work or releases together.

The reported outreach is not a public disclosure by OpenAI, a congressional decision, or an agreement among AI companies. The lawmakers involved and any response they gave have not been publicly established in the available reporting.

Why would a shared AI slowdown raise antitrust concerns?

A company can choose independently to slow its own research or postpone a product release. A shared pause is different because competing companies would need to agree on common limits, timing and compliance. Antitrust law broadly polices agreements among competitors that restrain competition.

Nicholas Felstead, an assistant director at the Australian Competition and Consumer Commission and former fellow at the Center for Law & AI Risk, wrote in March that a coordinated AI-development pause could be seen as restricting output and could potentially violate the Sherman Antitrust Act. He said the legal result would depend on the exact agreement, while uncertainty itself could discourage companies from working together.

That leaves an awkward trade-off for labs that favor collective safety measures. A unilateral pause gives competitors the option to continue advancing. But the coordination that could prevent one firm from racing ahead may itself create the antitrust issue.

What does OpenAI want to slow down?

The discussion concerns frontier AI, meaning the most capable models being developed at the leading labs. OpenAI chief scientist Jakub Pachocki recently argued publicly that the research field should coordinate to slow future development. He said voluntary slowdowns could become common until companies establish shared safety thresholds.

A shared safety threshold would be a common standard for deciding when a model’s capabilities or risks require more testing, monitoring or other safeguards before progress continues. Pachocki’s view does not establish that OpenAI has adopted a pause, or that competitors support one.

A bill offers a narrower path

In July, bipartisan lawmakers in both chambers introduced the Collaboration on Adversarial Threats and Security Risks Act. According to WIRED, the proposal would let AI labs cooperate on security and safety work without risking antitrust violations. The House bill was referred to the Judiciary Committee and had not been taken up.

The reporting does not establish that the proposal would authorize companies to agree on a broad slowdown in development. Its described purpose is narrower: allowing collaboration on safety and security, including sharing information and defensive measures. It has not been established as law or as a legal safe harbor for a collective pause.

Even with clearer legal rules, other barriers remain. WIRED reported that AI developers are competing for business and talent, some executives emphasize competition with China as a national-security concern, and major labs differ on how safe AI should be developed. John Schulman, an OpenAI cofounder now at Thinking Machines, argued that antitrust law does not prevent companies from jointly drafting a proposal, though it can prohibit particular agreements.

This story draws on original reporting from Decrypt.