For about three years, the complaint about AI companies has been the same one. They move too fast. They ship to the public and fix it afterwards. They let children be the test group.

On Friday, September 19, 2026, four people sued Anthropic, OpenAI, SpaceXAI and Google for the exact opposite.

A proposed class action was filed in the U.S. District Court for the Northern District of California. It alleges the four leading AI companies violated antitrust law by agreeing to coordinate a slowdown in AI development - and that doing so reduces the value consumers get from paid AI subscriptions.

The four named plaintiffs are not activists or rival firms. They are customers. They pay for ChatGPT, Claude, Grok or Gemini, and they are suing on behalf of a proposed nationwide class of everyone else who does.

The alleged conspiracy happened in public, on one day.

On September 12, Anthropic CEO Dario Amodei published an essay warning that swarms of rogue AI agents could take over the internet in as little as six months. He argued AI companies should move more cautiously, and proposed a three-point plan with the goal of "pacing the frontier."

The same day, per the lawsuit, his three biggest rivals - OpenAI's Sam Altman, SpaceXAI's Elon Musk and Google DeepMind's Demis Hassabis - each publicly responded in agreement.

That is the coordination. An essay and three replies.

The plaintiffs' case reduces to a single sentence:

Their lead attorney, Nick Rowley, put the stakes differently:

Note what that is actually saying. Rowley is not arguing AI is safe. He is arguing that safety decisions of this magnitude should not be made privately by the companies that profit from them.

And the plaintiffs are careful about something the headlines will flatten. Per the AP copy, they are not against these companies asking Congress, the White House or any other agency to develop AI regulation - and they are not even against the companies asking for an antitrust exemption. Their objection is to skipping that and agreeing among themselves.

Amodei raised the antitrust problem before anyone sued him.

In the original essay, he acknowledged the legal challenge directly, writing that it would help for the U.S. government to mediate "or at least enable" cross-lab discussions. The government wouldn't need to participate, he wrote, but would need to "issue a narrow waiver for certain kinds of safety conversations."

He asked for permission. He did not get it, and the lawsuit arrived instead.

Altman's response pointed the other way. He said OpenAI welcomes a "federal framework that sets consistent safety requirements" but added: "we do not believe we need to wait for an anti-trust exemption or legislation to begin the work of providing this confidence."

And the waiver itself looks like a dead end. Sen. Josh Hawley (R-Mo.) said in a recent Senate hearing that there is "no world" in which he would give "the most powerful companies in the history of the world" an exemption from antitrust law, arguing they could collude and stifle competition.

Meanwhile President Trump rejected the calls for regulation on social media, claimed efforts to limit the technology are part of a "conspiracy," and questioned why industry leaders would call for rules that he says would "drive them into oblivion and bankruptcy." On Saturday he said he is forming an AI task force and appointing an "AI czar," with scant detail.

So: the legislative door is contested, the self-organised door is being litigated, and all four companies declined to comment.

This next section is our analysis, not a finding in the case.

Safety in a competitive market is a collective-action problem. If one company slows down alone, it loses ground to the one that doesn't - and the cautious firm's restraint buys the public almost nothing, because the fast firm still ships. Unilateral safety is a tax you pay for a benefit you can't deliver.

The obvious answer is for competitors to move together.

That is also, roughly, the textbook description of what antitrust law exists to prevent. Competitors agreeing to restrain output is the paradigm case. The law does not have an exception that reads unless they were being responsible about it - and there are good reasons it doesn't, because "we did it for safety" is the most reusable excuse in the history of cartels.

Both things can be true at once:

There is a legitimate way through this, and everyone involved knows what it is. It is a regulator. It is Congress writing a standard that binds all of them equally, so that being careful isn't a competitive disadvantage and isn't a private deal either. That route is currently blocked by the politics described above.

Again: this is our connection, not the lawsuit's. No child and no child-safety provision appears anywhere in the filing as reported.

But look at the list of things parents have actually been asking AI companies for. Real age verification. Crisis detection that routes to a human. Limits on companion bots for minors. Memory a parent can see. Slower rollout of new capabilities to accounts held by teenagers.

Every single one of those costs a company speed.

Which means each one faces the same structure: any company that does it alone is slower than the company that doesn't, and a child using the faster product gets none of the benefit. The features that would most help families are precisely the ones a single firm has the least incentive to adopt by itself.

If coordinating on them becomes legally dangerous, the realistic outcome isn't safer AI or faster AI. It's each company deciding alone, quietly, with a lawyer in the room.

That is an argument for regulation, not for private agreements - which is, ironically, roughly where the plaintiffs end up too.

Don't take a side yet. No court has ruled on anything. This is a filing, and filings are arguments.

Watch what the companies do, not what they say. If safety commitments quietly get vaguer over the next few months, this lawsuit is a plausible reason.

Keep asking for the specific things. Age checks, crisis routing, parental visibility. Company-by-company pressure is not affected by this case, and it still works.

Support the boring route. Every serious version of this ends with a rule that applies to everyone equally. That is less satisfying than a lawsuit and considerably more useful.

The companies building this technology now face a choice with no clean option: race each other and be blamed for recklessness, or coordinate on caution and be sued for collusion.

That isn't a gotcha. It is a description of what happens when something this consequential is left to be worked out by competitors with no referee.

Source: Associated Press, "Lawsuit claims AI companies made an illegal deal to slow AI development," September 19, 2026. Accessed via CBS News and Fortune syndications of the AP wire copy. https://www.cbsnews.com/news/ai-slowdown-lawsuit-openai-anthropic-google/

All quotations verbatim from the AP copy. The lawsuit's claims are allegations; no court has ruled. The collective-action analysis, the child-safety connection and the guidance for parents are ours, not findings in the case.

Conflict of interest: This package concerns Anthropic, was drafted using Claude, and Claude is one of the four named subscription products in the lawsuit.

Want practical AI guidance for parents and educators every week? Subscribe: https://www.aibyage.com/?modal=signup&utm_source=beehiiv&utm_medium=newsletter