Anthropic, OpenAI, SpaceXAI and Google Sued Over an Alleged Pact to Slow AI Down

The AI safety debate just got a legal venue. On Friday, four people who pay for ChatGPT, Claude, Grok, or Gemini sued the four biggest AI companies in federal court — not over model safety, but over the opposite: an alleged agreement to slow down. The suit claims Anthropic, OpenAI, SpaceXAI (Elon Musk's AI company, the Grok maker), and Google coordinated to pace their development, and that a competitor agreement to slow down is exactly what antitrust law forbids. The safety conversation that started with an essay asking for a government antitrust waiver now faces a lawsuit arguing no waiver is needed — because the coordination was illegal all along. For builders, this is the clearest signal yet that the acceleration-versus-safety fight will be settled by courts and markets, not by essays.

What Actually Happened

The lawsuit was filed Friday in the U.S. District Court for the Northern District of California (CBS News). The four named plaintiffs — Cheyenne Hunt, Florida attorneys Charles Buist and Nick Spetsas, and California resident Christine Bullock — all pay for AI subscriptions, and they're seeking class-action status on behalf of a proposed nationwide class of paid subscribers to ChatGPT, Claude, Grok, and Gemini (The Epoch Times, AP).

The legal theory is consumer antitrust: the four biggest AI labs allegedly agreed to coordinate slowdown efforts, and doing so reduces the value customers get for paid subscriptions. "The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous," the plaintiffs argue (CBS News). Lead attorney Nick Rowley put the class's position more bluntly: "AI will quickly spin out of human control and could kill us all if we allow AI safety and protocol ... to be controlled by private self-serving agreements between the world's most powerful 'for profit' technology companies" (CBS News). Read that twice — the lawyers suing over the slowdown are not arguing AI is safe. They're arguing that even if it is dangerous, safety is a matter for regulators and law, not a private deal among rivals.

The coordination allegedly began on September 12, when Anthropic CEO Dario Amodei published an essay urging industrywide cooperation on decelerating advancement in favor of enhanced safety measures, warning that swarms of rogue AI agents could take over the internet in as little as six months (CBS News). His rivals — OpenAI's Sam Altman, SpaceXAI's Elon Musk, and Google DeepMind's Demis Hassabis — confirmed their agreement the same day, according to the complaint (CBS News); PBS NewsHour's account, cited by Hoodline, describes each publicly responding in agreement.

There's one carve-out that matters. The plaintiffs say they do not oppose the companies seeking AI regulation from Congress, the White House, or a federal agency, and they don't object to the firms pursuing a formal antitrust exemption. Their objection is to informal coordination that happened outside any such process (AP, via Hoodline). That distinction is the whole case: lobbying for permission is legal; a competitor pact pending permission is not.

Representatives for Anthropic, OpenAI, Google, and SpaceXAI did not respond to comment requests as of Saturday (CBS News, AP).

The Paper Trail: One Essay, Four CEOs, Nine Days

To understand why plaintiffs filed within a week of Amodei's essay (The Epoch Times), look at what the essay actually asked for. Amodei's September 12 post, covered here when it published, laid out a three-point plan: embedded independent evaluators inside frontier labs, coordination among the frontier labs on safety standards, and international cooperation on AI risks (Reuters, via Hoodline). He explicitly said he was not calling for a halt to model training — only that companies take adequate time to safeguard their models.

The legal mechanism was the revealing part. Amodei proposed that the U.S. government mediate, "or at least enable," these cross-lab discussions, and suggested Washington "issue a narrow waiver for certain kinds of safety conversations." Reuters reported he acknowledged the approach would likely require targeted antitrust exemptions in the United States (Reuters, via Hoodline). In other words: the author of the slowdown plan knew competitor coordination is illegal by default and asked the government to fix that first.

The suit takes the position that the labs didn't wait. The complaint cites the same-day responses from Altman, Musk, and Hassabis as evidence of the agreement (CBS News). Altman, for his part, drew a different line on the legal mechanics: he said OpenAI does not need to wait for an antitrust exemption or legislation to begin safety work, while welcoming a federal framework with consistent safety requirements (The Advocate, via Hoodline). That statement is likely to matter in court — it's Exhibit A for the defense that no agreement existed, only parallel public positions.

Amodei, meanwhile, gave an interview to CBS News Sunday Morning that doubles as context for why Anthropic went here at all. "I don't think I fully just appreciated what it would actually be like when the progress was as fast as it was," he told CBS's Jo Ling Kent. "If we build in the right way, I think the probability of something bad happening is very low. If we build in the wrong way, the probability of something bad happening is very high" (CBS News).

Why the Antitrust Theory Has Teeth — and Where It Gets Hard

Start with the harm theory, because it's cleaner than the politics. These are paid subscriptions. If you pay monthly for frontier-model access and the four frontier labs agree to slow down, the plaintiffs' argument runs, you get less model for the same money — and the coordination that caused it was an agreement among competitors, the category antitrust law treats most severely. Whether a court eventually accepts that framing is unknowable today. What's not debatable is that it converts a safety debate into a consumer damages claim, which gives plaintiffs standing, discovery rights, and a damages framework that a policy disagreement never would.

The second teeth: the coordination freezes out smaller labs. A pacing agreement among Anthropic, OpenAI, SpaceXAI, and Google is also an entry barrier — the incumbents agreeing not to race means nobody else gets to race past them. That's the argument Senator Josh Hawley used to say he would not agree to give major AI companies an antitrust exemption to collaborate, arguing such collaboration could enable collusion and stifle competition (Hoodline). Note what's happened to the political map: the most AI-skeptical Senate Republican and four subscription-paying plaintiffs now agree on the mechanism, if nothing else.

The third layer is the timing, and it's brutal for the labs. Their waiver request landed in a Washington that has decided it wants no part of AI limits: Trump rejected calls for regulation, called limiting efforts a conspiracy, and said strong AI rules would drive companies "into oblivion and bankruptcy" (AP, via Hoodline), while announcing an AI task force and czar with a no-new-constraints posture. Jensen Huang told CBS News the extinction discourse is a "doomsday narrative" not grounded in science, put the odds of AI ending the world by 2030 at "0%," and argued the labs' regulation requests are really asks to be relieved of existing liability law. The most important supplier to all four labs thinks their safety coordination is a pricing problem in disguise. So the labs requested a waiver from a government that sees no need for one, and now face a suit claiming the coordination they did without one was unlawful.

Where does the case get hard? Two places. First, plaintiffs must prove an actual agreement — the same-day public statements are suggestive, but public endorsements of a safety essay are not, by themselves, a contract. Altman's explicit statement that OpenAI needs no exemption to do safety work gives OpenAI a clean narrative. Second, antitrust litigation is measured in years; the realistic near-term cost to the labs is discovery — internal messages between rivals about pacing, safety commitments, and release timing. That's the part that should worry the labs more than the damages math.

What Builders Should Take From It

• Don't build your roadmap on a paced frontier. The lawsuit is a live test of whether the slowdown holds. If any of the four labs keeps shipping at full speed — and Huang's stance suggests at least their supplier hopes so — capability assumptions you made this month may expire fast.
• The waiver is dead; contracts are alive. With Trump opposing limits and Hawley opposing exemptions, a legislative antitrust waiver is a remote outcome. If you need safety commitments from model vendors, get them in your vendor agreements — SLAs, audit rights, incident-notification clauses — not from assumed industry norms.
• If you compete with the frontier labs, this cuts in your favor. Antitrust pressure structurally discourages incumbents from coordinating to hold capabilities back. The pacing ceiling Amodei proposed may never bind the smaller players — including yours.
• Watch for class certification and the motion to dismiss. Those two filings will tell you whether this case is a negotiating lever or a real constraint. Either way, expect the labs' public statements on safety coordination to get lawyer-reviewed from here on.
• Liability posture is unchanged — use it. Huang's point stands regardless of this suit: existing cybersecurity, unauthorized-access, and product-liability law applies to AI products today. Your contracts and SLAs are where your protection lives, and no pending regulation will replace them.

The safety debate wanted a waiver. It got a class action instead. Courts don't pace frontiers, but they do subpoena the people who claim they did.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.