The Pacing Pledge Meets Antitrust Law: Can AI Rivals Slow Down Together?

The Moment Everything Changed

On Friday, four people who pay for ChatGPT, Claude, Grok or Gemini filed a proposed class action in federal court in San Francisco accusing Anthropic, OpenAI, SpaceXAI and Google of illegally agreeing to slow down AI development. The alleged agreement is the very one the industry celebrated a week earlier as a rare act of collective responsibility. Seven days turned a safety pledge into a legal exhibit, and the plaintiffs’ central claim is that “collective restraint” among competitors is precisely what antitrust law forbids. The question that now hangs over the whole AI safety debate is uncomfortably simple: if the only way to slow down is to slow down together, is doing so together against the law?

Background

The pact traces back to September 12, when Anthropic chief executive Dario Amodei published “We Must Pace the Frontier,” and OpenAI’s Sam Altman, xAI’s Elon Musk and Google DeepMind’s Demis Hassabis each publicly signalled agreement the same day. Amodei’s essay did not only ask for goodwill; according to Gizmodo, it called on the US government to “mediate or at least enable” safety cooperation through a standards body or waivers. Three days later OpenAI’s Chris Lehane confirmed that OpenAI, Anthropic and Google DeepMind had been quietly talking about safety for weeks, and TechCrunch reported that the discussions were set in motion in July by Hassabis, who floated a US-led oversight body able to trigger industry-wide slowdowns. Lehane argued that the labs need no antitrust waiver to cooperate, pointing to the airline industry as a precedent for rivals sharing safety information, and said OpenAI backs a provision in the proposed FRONTIER Act that would force top labs to admit independent verification organizations.

What Happened

The backlash arrived almost as fast as the announcement. Speaking at Georgetown University on September 15, FTC Chairman Andrew Ferguson said that when companies are “simultaneously coming to Washington and asking for a host of regulations and an antitrust exemption, all of my alarm bells go off,” as quoted by Gizmodo. The next day, Semafor revealed that Senators Jim Banks and Adam Schiff had tried to slip a narrow AI antitrust exemption into the annual defense bill, modelled on the 2015 Cybersecurity Information Sharing Act and limited to sharing threat intelligence and defending against Chinese espionage and model distillation. Google welcomed it, while David Sacks, who has advised the Trump White House on AI, replied that companies have “a basic obligation to make your products safe” and should not “demand antitrust waivers or liability waivers in exchange for that.” The bill itself remains stuck after Senate Democrats blocked debate over the summer in a dispute about Trump’s Iran policy.

Meanwhile the public sector moved down a different road. On September 18, California Governor Gavin Newsom signed an executive order directing state officials to speed up independent oversight of frontier AI, including recommendations on requiring on-site independent auditors at frontier labs and creating a “kill switch” for frontier models whose effectiveness is verified on an ongoing basis by an independent organization, as the Governor’s office describes it. The order also widens the definition of a critical safety incident to cover loss-of-control events like the Hugging Face attack. The same day Anthropic named Accenture, through its Faculty AI unit, as its first embedded evaluator in a deal the two firms say will involve at least $1 billion over five years, according to TechCrunch. And then came the lawsuit.

Why It Matters

The complaint, reported by ABC7 San Francisco, does not object to any single lab choosing to move more slowly for safety reasons. Its target is the agreement itself, which the plaintiffs say used the September 12 essay and the same-day replies as the visible moment of coordination while alleging that the arrangement began months earlier. It also cites a July statement from lab employees acknowledging “intense competitive pressure not to unilaterally slow.” That quotation cuts both ways: it is the labs’ best argument that they cannot restrain themselves alone, and the plaintiffs’ best evidence that they needed to agree with each other to do it.

This is the structural bind at the heart of AI governance. Safety pauses only work if rivals do them together, yet competition law is wary of rivals coordinating on output, whatever their motives. There are three ways out, and each drew fire this week. Private agreements invite lawsuits like Friday’s, statutory exemptions draw the FTC’s suspicion of what Ferguson called incumbents insulating themselves from challenge, and public mandates such as California’s auditors and kill switch face an administration that, according to TechCrunch, has now had the president call AI safety concerns a Democratic hoax. Whichever route wins will decide who sets the speed limit on the most consequential technology of the decade.

Expert Perspectives

Critics have not been shy about questioning motives. Sacks wrote that people should “stop pretending the motivation to slow down is purely altruistic,” and added that the easiest way not to build superintelligence is for labs to agree not to build it, as reported by the San Juan Daily Star. Nvidia’s Jensen Huang asked, “What better way to create demand than to create a problem?”, suggesting labs stress security fears to sell security products, while Cohere chief executive Aidan Gomez branded the proposals “a cartel by any other name.” Even a supporter of some coordination, Microsoft’s Satya Nadella, cautioned that the effort “cannot be controlled by a handful of entities, but must have broad representation.”

The lawsuit’s lead attorney, Nick Rowley, framed the case in the language of safety rather than commerce, saying AI “will quickly spin out of human control and could kill us all if we allow AI safety” to be “controlled by private self-serving agreements between the world’s most powerful ‘for profit’ technology companies,” per ABC7 San Francisco. Newsom, for his part, offered the public-sector answer: “We’re not waiting to act – we’re going to speed up our work on substantial and responsible AI oversight before it’s too late.” Notice that almost everyone in this fight now claims to want stronger safety; they disagree over who is allowed to deliver it.

What to Watch

The first hard date is November 16, when California’s expert recommendations on on-site auditors and a kill switch are expected. Before that, watch how the four defendants respond to the complaint, since none had commented by the weekend, and whether they argue that safety coordination through independent verifiers is lawful without any waiver, as Lehane insists. Watch the defense bill too: if the narrow Banks-Schiff exemption survives, it would be the first statutory recognition that AI labs may coordinate on some risks, and if it dies, the labs will be left to test the airline analogy in court. Finally, watch whether embedded evaluators like Accenture become the industry’s compromise, because they deliver outside scrutiny without requiring any rival to promise anything to another.