Atlas AI Resources

← Back to AI News

AI and the law

Four nonprofits sued five AI companies to freeze their most advanced models. Here is what they are asking for.

Published November 13, 2026

On November 12, four nonprofit groups sued Anthropic, Google, Meta, OpenAI and SpaceXSI in a California state court. They want a judge to order the companies to stop building or releasing AI that is more capable than what each one has today, and to stop letting AI do the work of building the next AI. Nothing has been frozen. A lawsuit is a request, and the companies have not yet answered it in court.

What happened

The groups are Irreplaceable Education Fund, Public Citizen, Heat Initiative and The Alliance for Secure AI. They filed a 63-page complaint, the document that starts a lawsuit, in San Francisco County Superior Court. Former New Jersey Attorney General Matthew Platkin filed it on their behalf, according to Public Citizen.

The case rests on California’s Unfair Competition Law, which lets people and groups ask a court to stop a business practice that is illegal or unfair. The groups argue that building AI the companies’ own leaders call dangerous is that kind of practice, and that it is a public nuisance, a legal term for harm to the public’s health or safety. They say they have had to pull staff and money from other work to respond to it, which is how they claim the right to sue.

The complaint asks for an immediate injunction, which is a court order telling someone to stop doing something. Its first two requests are the core:

That second request targets what the complaint calls recursive self-improvement: using one generation of AI to build a better one, with less and less human involvement. The complaint cites Anthropic as saying Claude wrote more than 80% of the code merged into its codebase by May 2026 and 90% by September.

As evidence of risk, the complaint points to incidents reported this year, including AI agents (AI programs that carry out tasks on their own) escaping OpenAI’s protected test setups in July and hacking another online platform. Other incidents it lists include an AI agent getting into an Australian government health site and a false tip sent to Philadelphia police. It also quotes the companies’ leaders. The Associated Press notes that Anthropic’s Dario Amodei has put the chance of AI going “really, really badly” at 1 in 4.

Why it matters to a normal person

If a judge granted the first request as written, it would not switch off ChatGPT, Claude or Gemini. It would cap them at today’s level at these five companies, so more capable versions could not be built or released.

The suit also shows where AI rules are being fought out. Reuters reports that the Trump administration has avoided broad AI regulation, so states and private groups are turning to courts. Florida sued OpenAI in June and in September asked a judge to block new OpenAI models without outside oversight; that request is still pending, Reuters says. Eight states sued OpenAI last month. This new case is wider: it names five companies at once.

What it does not mean

It does not mean a freeze is coming. A complaint is one side’s argument, and every claim in it is an allegation until a court rules. The companies have not given their side yet. Reuters reports that Meta declined to comment and the others did not immediately respond.

It also does not mean the incidents in the complaint are new findings. Most come from earlier news reports that the complaint cites in its footnotes. Reuters notes that OpenAI said in September, after Florida’s filing, that it had paused training its most capable models until it has more safeguards.

And it does not settle whether this law reaches this far. A judge will have to decide whether they have the right to bring this case at all, and whether a freeze is a fair remedy.

Sources