The Cartel Pleads Safety: Four Subscribers Sue the AI Slowdown

On Friday, four AI subscribers — three of them attorneys — filed Buist v. Anthropic PBC in the Northern District of California, accusing Anthropic, OpenAI, xAI, and Google of violating Section 1 of the Sherman Act.¹ The alleged agreement is not minutes in a boardroom but eight days of public text: Dario Amodei’s September 12 essay calling for “industry-wide coordination” to “pace the frontier,” and the same-day endorsements — Elon Musk (“Dario is right”), Sam Altman (agreed), Demis Hassabis (“the right path forward”).¹ The proposed class is every US paid subscriber; the relief sought is an injunction against the coordinated slowdown and a declaration that it broke antitrust law — no damages figure yet, just an order to stop being worse on purpose.² Plaintiffs’ lawyer Nick Rowley framed the target as “private self-serving agreements” that would let AI “quickly spin out of human control” without legislative scrutiny.³ Saturday brought the second answer to the essay: President Trump announced an “AI Force” modeled on Space Force, promised an AI czar — “Only High I.Q. individuals need apply” — called the labs’ safety warnings a “SICK conspiracy,” and pledged not to “hinder or stifle” an industry he sizes at a quarter of GDP.⁴ The essay that asked to slow the frontier got, inside four days, a subpoena and a czar. Four frames on what just happened.

Analogy Transfer — the stabilization defense

Strip the nouns and the structure is familiar: rivals in an essential industry coordinate to restrict output, publicly, citing a higher purpose. The structural twin with the longest track record is OPEC — oil ministers announcing quotas to “stabilize markets,” consumer governments answering with lawsuits. The mechanism that does the work there: once output restraint among competitors is public and explicit, motive stops mattering — the restraint itself is the violation. Translate back: under the Sherman Act, a safety motive is not obviously a defense to alleged output restriction, and the plaintiffs’ theory leans on exactly that — competitors agreeing to hold their product back.¹ Disanalogy check, because the analogy must earn its keep: OPEC restricts to raise prices for revenue; the labs claim to restrict to reduce risk, and restraints defended as quality-protecting have sometimes received gentler rule-of-reason review instead of near-per-se condemnation. But the plaintiffs’ injury form — subscribers alleging the product they pay for was deliberately kept worse — is the injury antitrust exists to recognize. The analogy holds where it matters: the restraint is the evidence, and the labs published it themselves.¹²

Question Forge — which question is on trial

The suit announces itself as answering “should AI development be slowed?” It will answer no such thing. The question a court can actually hear is narrower: did four competitors agree to restrain trade? A yes proves collusion, not recklessness; a no proves procedure, not safety. The displacement runs the other direction too — Trump’s announcement forges “who governs AI?” into “who is the czar?”, a question answerable by a single appointment to an office of unstated form; nobody will yet say whether the AI Force is a military command or a civilian agency.⁵ The forged question worth carrying is this: when a slowdown would benefit everyone but is priced by courts as harm to identifiable subscribers, who is authorized to weigh the whole against the part? Friday did not answer it. Friday made it harder to dodge.

Decision Matrix — the labs’ remaining moves

Rank the labs’ options on legal exposure, safety value, legitimacy, and reversibility. Public coordination — essays and endorsements — now supplies exhibits to a class action: maximal visibility, collapsing legitimacy. Private coordination is worse on every axis; secrecy converts a thin claim built on public statements into a plausible conspiracy with discovery rights attached. Racing — restoring the status quo the pacing essay was written to prevent — carries zero legal exposure and maximal safety cost by the labs’ own account. Legislative delegation — get the state to mandate pacing — converts collusion into compliance: joint petitioning of government is the classic antitrust safe harbor, and state-compelled conduct earns state-action immunity. The matrix ranks delegation first on every criterion except one: its gatekeeper. Sensitivity check: the winner flips to racing as soon as you weight speed-to-decision at all heavily, because delegation requires signature from a president who compares AI risk to the climate hoax and whose last AI czar was a venture capitalist who has since stepped down.⁴⁶ The lawsuit did not settle whether pacing is wise. It removed every legal path to pacing except the political one — and the political branch is currently held by a man who disbelieves the premise.

Ladder of Abstraction — from a Friday filing to the pattern

Bottom rung: four named plaintiffs, a Friday filing in San Francisco, a Saturday social post.¹⁴ Middle rung: the pattern — when an industry’s coordination benefits everyone but costs identifiable customers something measurable, antitrust hands those customers a lawsuit; and when the state declines to govern, private governance emerges, and private governance among competitors has another name. Top rung: externalities that are global (catastrophic risk, borne by all) paired with injuries that are individual (a suppressed product, borne by subscribers) cannot be reconciled by private contract between the parties causing both. They require a sovereign. Pacing was never technically hard. It is politically unowned.

Synthesis — the race, with a lawsuit attached

Read together, the frames converge on one finding: the suit is not a verdict on safety but a verdict on venue. Public coordination is dead — it is now evidence. Private coordination is worse. Racing is legal and, by the labs’ own stated beliefs, the dangerous default. Legislation is the only surviving road, and it runs through a president who calls the destination a hoax — days before he hosts Xi Jinping, with Jensen Huang cast as the industry’s emissary and Altman reportedly invited, at Thursday’s state dinner.⁷ The default outcome is the race continuing, with a class action now attached to every press release. The subscribers suing to make their products better may well win — and the pacing debate moves to the only forum that was never staffed for it.

1. https://thehill.com/policy/technology/6099571-lawsuit-accuses-anthropic-openai-spacexai-google-of-ai-pacing-collusion ↩ ↩² ↩³ ↩⁴ ↩⁵

2. https://cryptobriefing.com/users-sue-openai-anthropic-xai-collusion ↩ ↩²

3. https://www.tribuneindia.com/news/business/anthropic-openai-spacexai-google-face-federal-antitrust-lawsuit-over-calls-to-slowdown-ai-development/amp ↩

4. https://www.cnn.com/2026/09/19/politics/trump-ai-task-force-czar ↩ ↩² ↩³

5. https://www.nytimes.com/2026/09/19/us/politics/trump-ai-force.html ↩

6. https://www.nbcnews.com/politics/white-house/artificial-intelligence-task-force-czar-technology-trump-rcna598688 ↩

7. https://www.cnbc.com/2026/09/20/nvidia-ceo-jensen-huang-emerges-as-trumps-top-ally-in-ai-debate.html ↩