AI Labs Want a Slowdown Pact and an Antitrust Shield Too

The headline from four AI labs this fall was "we want to slow down for safety." The headline regulators should be reading is the paragraph underneath it: those same labs also want Washington to grant them legal cover for coordinating with each other, something antitrust law normally treats as a cartel. FTC Chairman Andrew Ferguson has a name for pairing those two requests. He calls it moat digging, and a private lawsuit filed in September argues the informal version of that arrangement already happened.

What the Four Labs Actually Signed Up To

On September 12, 2026, Anthropic CEO Dario Amodei published an essay, reported by Daily Journal as titled "We Must Pace the Frontier," arguing that the industry should coordinate on slowing the rate of frontier AI development rather than racing unilaterally. Within hours, OpenAI's Sam Altman, SpaceXAI's Elon Musk and Google DeepMind's Demis Hassabis had each publicly endorsed some version of that call, according to reporting from PBS NewsHour and TechSpot. None of that is illegal on its own. A chief executive can write an essay, and rivals can agree with it in public.

What turns opinion into a legal question, according to the complaint covered by those outlets, is a July 2026 statement signed by senior employees across several labs that acknowledged "intense competitive pressure not to unilaterally slow" development and called for government support of a coordinated, industrywide pacing effort. That is the detail the lawsuit leans on: not that any one company chose caution, but that rivals allegedly discussed doing it together so that caution would not cost any of them market share.

"Moat Digging" Is the FTC Chairman's Word for It, and It Fits

Andrew Ferguson, chair of the Federal Trade Commission, said in mid-September 2026 that he is "deeply suspicious" of AI companies that show up in Washington asking for both new regulation and an antitrust exemption at the same time, a combination he summarized as moat digging, according to Global Competition Review and Bloomberg Law. His reasoning is straightforward: regulation raises the cost of entering a market, and an antitrust exemption removes the one legal tool smaller rivals and enforcers have to stop incumbents from using that cost as a weapon. Put the two together and the labs asking loudest for "responsible pacing" are also the four companies best positioned to survive the resulting compliance burden.

That is not a hypothetical concern anymore. Law360 reported on October 7, 2026 that AI companies have moved from public essays to actively lobbying Congress and federal regulators for a formal antitrust carve-out that would let them coordinate a safety-driven slowdown without exposure to Sherman Act claims. If granted, that exemption would do exactly what Ferguson warned about: let the firms that already have the most capital, the most compute and the most customers set the industry's pace, while a formal legal shield keeps challengers from arguing the pace was set collusively rather than competitively.

A Private Lawsuit Already Treats the Quiet Version as a Cartel

Four paid subscribers to ChatGPT, Claude, Grok and Gemini filed suit on September 18, 2026 in the US District Court for the Northern District of California, naming Anthropic, OpenAI, Google and SpaceXAI as defendants. The case, filed under Section 1 of the Sherman Act, does not argue that any one company slowing its own product is illegal. It argues that four competitors agreeing, even informally and in public, to slow down together is a classic output-restricting arrangement, and that subscribers paying monthly fees for frontier AI access were the ones who absorbed the cost of slower progress. The complaint seeks treble damages, an injunction against the alleged arrangement and certification of a nationwide class, per reporting from TechSpot and Daily Journal.

The lawsuit's evidence is entirely public: Amodei's essay, the same-day endorsements, and the July employee statement. One Daily Journal analysis of the filing noted it contains no internal emails, no confidential witnesses and no meeting minutes, a point the defendants will almost certainly raise if they move to dismiss. Proving that four rivals coordinated based only on things they said out loud is a harder case than proving it from a leaked memo, and courts have historically been reluctant to treat parallel public statements alone as proof of an illegal agreement.

The Steelman: Safety Coordination Has a Real Legal Precedent

The strongest case against treating this as pure moat digging is that competitors coordinating on safety is not new, and regulators have blessed it before. A 2014 FTC and Department of Justice policy statement held that properly designed sharing of cybersecurity threat information among competitors generally does not raise antitrust concerns, because the public benefit of faster threat response outweighed the competitive harm of firms talking to each other, according to legal analysis cited by Mondaq and Wiley. If AI labs limited themselves to sharing safety research and incident data the way security teams already share breach information, that would look a lot more like the 2014 framework than like price fixing.

The problem is that the current ask goes further than information sharing. Amodei's call and the July statement are not about exchanging safety research, they are about agreeing on the pace of capability releases, which is the one dimension these companies actually compete on. Anthropic's own pricing moves this year show how closely pace and competitive position are linked, and a pacing agreement among the four largest labs would functionally cap how fast smaller rivals and open-source developers could catch up, since the leaders would no longer need to outrun each other to stay ahead of everyone else. That is a materially different ask than sharing a list of known malware signatures.

Who Should Actually Be Watching This

This story matters most to developers, startups and enterprise buyers who depend on frontier model access and to anyone tracking AI policy, since the outcome will shape how fast competing models improve and how much leverage smaller players retain. It matters less to casual consumers choosing between chatbot apps this week, since nothing here changes current pricing or features, and the case could take years to resolve or be dismissed outright. Regulatory scrutiny of these labs has been building on multiple fronts, and this lawsuit is one more data point in that pattern rather than an isolated event.

• Watch for a motion to dismiss in Buist v. Anthropic PBC: if the court rules that public statements alone cannot establish a Sherman Act agreement, the case likely dies early and the labs' informal coordination continues unchecked.
• Watch Congress and the FTC for any movement on a formal antitrust exemption: a narrow waiver limited to safety research sharing is a different outcome than a broad exemption covering pacing and release timing.
• If you build on these models commercially, treat any future joint "pacing" announcement from two or more labs as a signal to diversify vendors rather than a reassurance about safety, since a slower incumbent pace benefits incumbents more than it benefits you.
• Skip the day-to-day headline chasing on this one; the legal process moves in months, not days, so checking in quarterly is enough unless you are directly affected by procurement decisions.

The Honest Take

Safety concerns about frontier AI are real, and it would be wrong to assume every word of Amodei's essay was cynical. But intent is not the legal question, and it is not the competitive question either. Even a genuinely safety-motivated agreement among the four best-funded labs to slow down together still freezes the current market order in place, because it removes the one mechanism, unilateral competition, that has historically let challengers close a gap. Ferguson's instinct that regulation plus exemption equals moat digging does not require assuming bad faith from Amodei, Altman, Musk or Hassabis individually. It only requires noticing that the policy outcome they are asking for would benefit exactly the four companies asking for it, regardless of why they are asking.

Sources

• PBS NewsHour: Lawsuit says Anthropic, OpenAI, SpaceXAI and Google made illegal agreement on AI slowdown
• TechSpot: Anthropic, OpenAI, Google and SpaceXAI face lawsuit claiming AI slowdown harms subscribers
• Daily Journal: AI giants sued over alleged pact to slow development
• Bloomberg Law: OpenAI, Anthropic, Google, SpaceXAI Hit With Antitrust Lawsuit
• Global Competition Review: Ferguson: AI regulation risks "moat digging"
• Mondaq: AI Safety Collaboration and Antitrust, lessons from recent enforcer statements