The Legal Loophole in AI Slowdowns: When Safety Meets Antitrust

The Question at Hand

When OpenAI recently asked Congress whether it's legal to coordinate an industry-wide slowdown in AI development, it wasn't just seeking policy guidance. It was grappling with a legal gray area that could define the future of artificial intelligence regulation. The question touches on a fundamental tension: Can AI companies collaborate to ensure safety without violating antitrust laws?

At first glance, this seems like a straightforward issue—collaboration for public safety is usually welcomed. But as we dig deeper, the answer becomes anything but simple. This is where the rubber meets the road between ethics and enforcement.

A Shift in AI Ethics

The idea that AI systems could become self-improving and pose existential risks has moved from theoretical concern to a real fear among researchers and policymakers alike. In a recent blog post, OpenAI's chief scientist Jakub Pachocki argued that voluntary slowdowns are necessary until shared safety standards are in place.

"The best path forward includes coordinating to slow down future development," he wrote. "Voluntary slowdowns to become commonplace until shared safety bars are established."

This perspective reflects a growing sentiment in the AI community: that safety must come before speed. But for many, the real question isn't whether such coordination is wise—it's whether it's legally permissible.

The Antitrust Challenge

Legal experts point to the Sherman Antitrust Act as a potential stumbling block. The law prohibits agreements that restrict trade or competition, which could include any kind of coordinated output limitations in the AI sector.

Nicholas Felstead, a former AI policy fellow at the Center for Law & AI Risk, wrote that such coordination might be seen as an attempt to control market dynamics, even if the intent is safety-focused. He emphasized the need for precise legal frameworks that distinguish between harmful collusion and responsible cooperation.

"Legal uncertainty can act as a powerful deterrent," Felstead noted in a March article. "Even if most safety collaborations would ultimately survive antitrust scrutiny, the risk of prosecution alone may be enough to halt development."

Congressional Response

Lawmakers appear to be taking notice. In July, a bipartisan group introduced the "Collaboration on Adversarial Threats and Security Risks Act," which explicitly permits AI labs to coordinate on safety efforts without antitrust consequences.

The bill has been referred to the House Judiciary Committee but remains in the early stages of legislative review. Caleb Knapp, director of government affairs at the AI Policy Network, supports it as a critical step toward a safer AI future.

"We need legal channels for industry to work together," he said. "The risk of violating antitrust law is holding back important safety discussions."

Industry Hesitations

However, even with a legislative path forward, industry resistance persists. Some executives argue that concerns about antitrust are just cover for deeper strategic issues.

"They'll cite antitrust, but that's fake," said John Schulman, a cofounder of OpenAI now at Thinking Machines. "Antitrust prohibits certain agreements, but not from jointly developing a proposal."

Indeed, the AI landscape is fiercely competitive. Companies like OpenAI and Anthropic are locked in an intense race to build the next generation of frontier models. Beyond business rivalry, there are fundamental disagreements over how AI should be developed—especially regarding safety protocols.

The Human Element

That tension became clear this summer when former researchers from both companies raised alarms. Jacob Coxon, a former Anthropic researcher, warned that AI developers were putting humanity at risk.

Security incidents, such as OpenAI's agents hacking Hugging Face, have only intensified those concerns. As the industry struggles to keep pace with rapid advancements, the urgency to regulate grows louder—and more urgent.

What Comes Next?

The path ahead is uncertain but critical. If AI labs can't legally coordinate safety efforts, we risk a dangerous divide between innovation and responsibility. The stakes are high: the decisions made now will shape how artificial intelligence evolves for years to come.

We must find ways to encourage collaboration while safeguarding fair competition. Only then can we hope to build an AI ecosystem that's both powerful and safe—a future worth striving for, not just a technological marvel.