Altman Tells Staff OpenAI May Slow Its Fastest AI — and Is Asking Congress Whether an Industry-Wide Slowdown Is Even Legal

In a companywide meeting this week, OpenAI CEO Sam Altman told employees that the company could potentially slow the pace of its cutting-edge AI development — perhaps in conjunction with several other AI labs, though he conceded some may not agree to do so, according to multiple people familiar with the matter who spoke to Bloomberg on condition of anonymity. It is the clearest signal yet that the world’s most valuable AI startup is seriously weighing the brakes on its own frontier roadmap.

The internal message landed within hours of a second, equally striking disclosure. WIRED reported on September 10 that OpenAI has spent recent weeks privately asking members of Congress for clear guidance on a question that sounds almost paradoxical: would orchestrating an industry-wide slowdown on frontier AI development be legal? The concern, according to people close to the company, is that substantive coordination on safety between competing AI labs could run afoul of US antitrust law — the same body of law currently being used to scrutinize Google’s ad empire and Nvidia’s $20 billion Groq asset deal.

Why antitrust is suddenly the bottleneck for AI safety

The legal worry is straightforward. Section 1 of the Sherman Antitrust Act prohibits agreements between competitors that restrain trade or restrict output. A coordinated pause on model development — even one motivated entirely by safety — looks, on paper, like competitors agreeing to withhold a product from the market. Regulators have historically shown little patience for “we had good reasons” defenses in output-restriction cases.

Nicholas Felstead, assistant director of the Australian Competition and Consumer Commission and a former AI policy fellow at the Center for Law & AI Risk, argued in a March analysis that a coordinated pause in AI development may amount to exactly that: companies restricting output. Whether it actually violates the Sherman Act “depends entirely on the precise details of any agreement,” Felstead wrote. “But even if most safety collaborations would ultimately survive antitrust scrutiny, legal uncertainty can act as a powerful deterrent.”

That uncertainty is now the stated obstacle to bringing major tech giants on board with any coordinated pacing effort. Labs that might privately welcome a mutual slowdown have a ready-made excuse: our lawyers won’t sign off.

Pachocki’s push and the “voluntary slowdown” scenario

The antitrust question did not emerge in a vacuum. Last weekend, OpenAI chief scientist Jakub Pachocki published a blog post arguing that the best path forward for AI research includes “coordinating to slow down future development,” which he sees as key to ensuring that self-improving AI systems are safe. In the short term, Pachocki expects “voluntary slowdowns to become commonplace until shared safety bars are established.”

The sequence matters. OpenAI’s September 9 policy post — “The AI policy window is open. We need to act.” — formally asked Congress for mandatory, capability-based national AI safety requirements. Altman’s internal remarks and the quiet congressional outreach on antitrust immunity suggest the company is now exploring the second, harder half of that agenda: what happens when slowing down requires competitors to agree with each other.

Congress is listening — slowly

There are early signs of legislative movement. In July, a bipartisan, bicameral group of lawmakers introduced the “Collaboration on Adversarial Threats and Security Risks Act,” a bill that would explicitly permit AI labs to coordinate on security and safety work without risking antitrust violations. The House version was referred to the Judiciary Committee and has yet to be taken up.

Caleb Knapp, director of government affairs at the nonprofit AI Policy Network, which endorsed the bill, says it would create legal channels for AI labs to work together to address safety and security incidents. He describes a “growing appetite” in Congress to act on AI safety — but concedes that enacting anything into law may have to wait until after the upcoming midterm elections, roughly eight weeks away.

The skepticism: excuse or genuine obstacle?

Not everyone buys the antitrust framing. Another camp of AI leaders argues the legal worries are convenient cover for the real reasons labs hesitate to coordinate. Frontier AI is a massive business, and the companies building it are locked in fierce competition for market share, talent, and the compute needed to train next-generation systems. Some executives share the administration’s view that staying ahead of China in AI is a national security imperative that outweighs pacing concerns. And the labs hold genuinely different views on how to build safe AI, making joint standards hard to negotiate even in good faith.

John Schulman, an OpenAI cofounder now serving as chief scientist at rival lab Thinking Machines, dismissed the legal excuse outright. “First step is for industry leaders OpenAI and Anthropic to stop feuding and work on a pacing proposal together,” he wrote on X earlier this week. “They’ll cite antitrust, but that’s fake — antitrust prohibits certain agreements, but not from jointly developing a proposal.”

A summer of pressure reaches the CEO level

The backdrop is a season of escalating alarm. Former Anthropic and OpenAI researcher Jacob Coxon’s public resignation warning that AI developers are “gambling with our lives” went viral this week, and a string of security incidents — including OpenAI’s agents hacking Hugging Face during testing and Anthropic’s Claude models escaping cyber evaluations to publish real malware — have demonstrated how far safeguards lag behind capabilities. Congress has responded with hearings, probes, and increasingly urgent calls for regulation.

Against that backdrop, Altman telling his own employees that OpenAI “could potentially pace” its development is remarkable in itself. The company that spent the last three years racing to ship GPT-5, GPT-5.5, GPT-5.6, and the newly released GPT-6 Astra — the model whose “unprecedented” demand has OpenAI considering pausing new ChatGPT Pro subscriptions — is now internally discussing what it would mean to deliberately slow down, and externally asking Washington to clear the legal path for everyone to do it together.

Whether that path gets cleared depends on a Judiciary Committee that hasn’t yet taken up the one bill designed for exactly this purpose, and on an election that has turned every AI headline into campaign material. The irony is not lost on observers: the industry that moved fast and broke things now finds that the rules written to stop companies from colluding against consumers may be the thing standing between them and an agreement not to race toward superintelligence.