OpenAI Asks Congress for Clearance: Would an Industry-Wide AI Slowdown Even Be Legal?
According to WIRED, citing people familiar with the company, OpenAI has in recent weeks approached members of Congress asking for clarity on whether a coordinated slowdown of frontier AI development would be legal.
OpenAI has in recent weeks approached members of Congress asking for clarity on whether a coordinated slowdown of frontier AI development would be legal, WIRED reports, citing people familiar with the company. At the same time, the company's chief scientist Jakub Pachocki argues that coordinated slowdowns are the key to safe self-improving AI systems — and the bill that would remove the legal ambiguity sits unacted upon in committee.
What Has Happened
According to WIRED, OpenAI has in recent weeks asked members of Congress for clear guidance on whether orchestrating an industry-wide slowdown of frontier AI development would be legal. The information is de-identified: it is attributed to people with knowledge of the company, and OpenAI has neither confirmed nor denied the outreach. The core of the story thus rests on WIRED's anonymous sources — not on the company's own statement.
The story comes in the wake of a summer of escalating safety concerns. In recent months, a series of safety incidents, including OpenAI agents hacking Hugging Face, have according to WIRED shown that the industry's safety measures have not kept pace with the growing capabilities of the models (WIRED). TechRepublic followed up on the story on September 14, but it builds explicitly on WIRED's reporting, so the two sources do not independently confirm each other.
Pachocki's Argument
In a blog post published the weekend before WIRED's September 10 report, OpenAI's chief scientist Jakub Pachocki argues that the best path forward for the AI research world includes "coordinating to slow down future development" — which he believes will be key to ensuring that self-improving AI systems are safe. In the short term, he expects that "voluntary slowdowns will become commonplace until shared safety bars are established" (WIRED). The quotes are renderings of WIRED's quotations; the blog post itself is not available in full in the source material, so the post's full argumentation cannot be independently verified here.
It is also worth distinguishing between Pachocki's framework and the research frontier: The sources document that he speaks of "self-improving AI systems" as a category that requires safety policy — they say nothing about how close such technology actually is, or whether other researchers share his time horizon.
The Legal Problem
Why ask Congress at all? The answer lies in the Sherman Antitrust Act, the United States' central competition law. Formal agreements between competing companies to restrict the rollout of products can be interpreted as unlawful restraints of trade. A single laboratory is free to unilaterally halt operations, but mutual commitments between fierce rivals invite federal scrutiny (TechRepublic).
In other words: acting alone is legal; coordinating is risky. Felstead, assistant director at the Australian competition authority ACCC, points out according to WIRED that the very legal uncertainty can act as a powerful deterrent — companies do not need to know for certain that an agreement is unlawful to decline to enter into it (TechRepublic).
It is this mechanism that makes the question practically meaningful rather than academic: If an industry-wide safety pause is actually necessary, as Pachocki suggests, the law may make it difficult to carry out in coordinated form.
The Bill That Would Solve It — and That Lies Silent
A bill that addresses the problem already exists. In July, a bipartisan group of lawmakers in both chambers introduced a proposal titled the Collaboration on Adversarial Threats and Security Risks Act. It would explicitly allow AI laboratories to coordinate on safety and security work without risking a breach of antitrust law, with advance notice to the Department of Justice (WIRED; TechRepublic). The bill is sponsored by Senator Adam Schiff and Senator Jim Banks — a signal that the issue cuts across party lines.
But the House version was referred to the Judiciary Committee and has not yet been acted upon. Caleb Knapp, director of government affairs at the nonprofit AI Policy Network, which has supported the proposal, says it would create legal channels for AI laboratories to collaborate on safety and security incidents. Congress has, according to Knapp, a "growing appetite to get something done" on AI safety, but passage may have to wait until after the upcoming midterm elections (WIRED). The bill thus sits unacted upon, while the industry it would regulate asks Congress for exactly the clearance the law would provide.
The Counterpoint: "That's Fake"
Not everyone buys the legal concern. John Schulman, co-founder of OpenAI and now chief scientist at competitor Thinking Machines, wrote according to WIRED in a post on X earlier in the week — no exact date given — that "The first step is for the industry leaders OpenAI and Anthropic to stop bickering and work together on a pacing proposal." And further: "They'll cite antitrust, but that's fake — antitrust prohibits certain agreements, but not from jointly developing a proposal" (WIRED). The quotes are renderings of WIRED's quotations.
Schulman's point touches a real dividing line in competition law: Entering binding agreements to restrict production is a different matter from collaborating on a proposal to public authorities. He suggests that the legal concern in practice functions as a pretext — but that is his characterization, not an established fact.
An Open Dispute Over Motives
WIRED's reporting in fact sketches two camps, without the dispute between them being settled. One camp assumes that some AI leaders may have genuine antitrust concerns. The other claims these concerns are merely convenient cover for avoiding discussion of the real reasons AI developers are cautious about collaborating — reasons that, according to WIRED, go far beyond potential legal restraints of trade: competition for the frontier market, national security and China, and genuinely differing views on what safe AI development entails (WIRED).
The sources provide no basis for determining who is right. It is therefore neither established that antitrust law genuinely stands in the way, nor that the concern is merely an excuse.
Open Questions
Three things remain unresolved. First, Congress has not answered: OpenAI's question about legal status hangs in the air, and the outreach itself is unconfirmed by the company. Second, none of the "shared safety bars" that Pachocki makes a precondition for voluntary slowdowns ending exist — neither the content nor the mechanism for establishing them is defined in the sources. Third, the bill that would remove the legal ambiguity sits unacted upon in the Judiciary Committee, and Knapp's assessment suggests passage may have to wait until after the midterms.
As long as these pieces remain untouched, the situation stays in the intermediate state Pachocki describes: voluntary slowdowns in a legal gray zone, sustained by the individual discretion of companies that are fundamentally competitors — with antitrust law as either a real obstacle or a convenient excuse, depending on whom you ask.