There’s a question people rarely ask about “the labs should slow down together”: are they allowed to? Two companies agreeing to ship a product later is, in most industries, called a cartel.
According to WIRED, OpenAI has now put exactly that question to members of Congress. The company wants to know whether an industry-wide coordinated slowdown in AI development would be lawful under US antitrust law. The worry is that coordinating with other labs on safety could violate the Sherman Antitrust Act.
How it got here
Sam Altman told staff this week that OpenAI could slow its pace, possibly alongside other labs — though some probably wouldn’t go along. Chief scientist Jakub Pachocki argued in a blog post for coordinating a slowdown until shared safety standards exist.
The trigger is a run of incidents, including the OpenAI agents that attacked a third-party website. It also includes accusations from a former OpenAI and Anthropic employee that the industry is gambling with humanity’s survival, which broke into mainstream coverage over the past few days.
The direction isn’t new. Back in July, more than 1,000 employees at major AI firms signed a petition calling for a mechanism to slow development.
The bill that would fix it has been in committee since July
There is already a draft aimed at the antitrust problem: the bipartisan “Collaboration on Adversarial Threats and Security Risks Act”, which would let labs work together on safety issues. Introduced in July, sitting with the Judiciary Committee ever since.
Put differently: the answer to OpenAI’s question exists as text, just not as law.
My take
The antitrust argument only half convinces me. The Sherman Act targets agreements on prices and output, not every form of joint safety work — industry standards exist in aviation, pharma, and most other regulated sectors. Asking the question anyway looks less like law than like positioning: saying publicly “we would, but we’re not allowed to” moves the responsibility toward Congress.
Both can be true at once, though. The legal uncertainty is real, and a company that wants to coordinate would sensibly get cover first. And a bill parked in committee since July is exactly the kind of blockage a public inquiry can shake loose.
The timing is what stays with me. In the same week, the Senate is negotiating a duty of care that would let the government hold models back. If that arrives, nobody needs an agreement between labs at all.
Sources: The Decoder: OpenAI floats a shared AI slowdown, takes it to Congress, Decrypt: OpenAI Asks Congress Whether an AI Slowdown Would Be Legal, Techmeme roundup