OpenAI Seeks Congress’s View on a Coordinated AI Slowdown

A voluntary pact among leading AI labs to slow development could face a legal barrier before it faces a technical one. OpenAI has asked members of Congress for guidance on whether a coordinated industry-wide slowdown in frontier-AI development would be lawful, according to people close to the company.

The request puts a sharp constraint on a growing safety argument. OpenAI has publicly said it wants to work with other frontier labs on voluntary standards, including shared standards for when development should slow or stop. Its policy position is that safety requirements should set the pace when capability growth cannot meet necessary safety bars.

A safety proposal that can resemble market coordination

The legal concern is not abstract. Legal scholars have warned that an agreement to limit AI development might be treated as a restriction on output under US antitrust law. Whether it would violate the law would depend on the agreement’s exact terms, but uncertainty alone can deter companies from joining a safety arrangement.

That creates an awkward split between setting a common safety goal and agreeing on conduct that achieves it. OpenAI chief scientist Jakub Pachocki has argued that research organizations should coordinate to slow future development, expecting voluntary slowdowns to become common until shared safety bars are in place.

Congress has a narrow proposal, not a settled answer

Lawmakers have already introduced one possible route around the problem. A bipartisan, bicameral group introduced the Collaboration on Adversarial Threats and Security Risks Act in July. The bill would allow AI labs to coordinate on safety and security work without risking antitrust liability.

But the measure is not yet a solution available to companies. Its House version was referred to the Judiciary Committee and had not been taken up when WIRED published its report. OpenAI has separately called for mandatory, capability-based national AI safety regulation, while saying industry standards should complement rather than replace federal safeguards.

Agreement on a proposal is not agreement on a pause

The antitrust issue may be narrower than it first appears. John Schulman, an OpenAI cofounder now serving as chief scientist at Thinking Machines, argued that OpenAI and Anthropic could jointly develop a pacing proposal even if some agreements between competitors are prohibited. His point was that a proposal and a binding arrangement to restrict development are different acts.

The choices now in view

• Labs could develop voluntary standards while avoiding commitments that could be seen as restricting output.
• Congress could create a legal safe harbor for specified safety and security collaboration.
• Federal rules could establish mandatory safety requirements instead of leaving pacing solely to private agreements.

OpenAI’s outreach does not establish that a coordinated slowdown would be illegal, or that rival labs would agree to one. It does show that the company sees the law as a potential obstacle to the collective safety system it is urging. The unresolved question is whether Washington will define a permitted path before private competitors attempt to define one themselves.