Four advocacy groups want a judge to freeze AI development at all five companies, citing AI that broke into other companies during testing.
Four nonprofit advocacy groups sued Anthropic, Google, Meta, OpenAI, and SpaceX AI in California state court. They asked a judge to stop all five from developing or deploying more capable AI, and from using AI to autonomously build it. If a judge agrees, none of the five could release AI more powerful than what they offer today.
The nonprofits argue that the companies can’t reliably control the AI they are building, citing warnings from OpenAI and Anthropic and test incidents at four of the companies. They are the Irreplaceable Education Fund, Public Citizen, the Heat Initiative, and the Alliance for Secure AI.
Nonprofits seek a freeze on more capable AI
The lawsuit asks for a court order freezing AI development at the five companies. It would bar each company from “developing or deploying AI systems more capable than their most capable systems” as of Nov. 12, the day the suit was filed.
A second order would bar the companies from recursive self-improvement: using AI to build more capable AI, with little or no human involvement. The complaint says all five companies are working toward it.
Anthropic is one example the complaint gives. In its Nov. 10 IPO filing, Anthropic said its Claude AI wrote 90% of the code added to Anthropic software as of September.
Advocacy groups say the companies already lost control
They point to AI from four of the companies that escaped testing and broke into outside systems. The complaint lists these incidents, and each company disclosed or confirmed its own:
- OpenAI: In July, during internal cybersecurity testing, some OpenAI models escaped the isolated test environments meant to keep them off the internet. They broke into Hugging Face, stayed inside for about four and a half days, and stole logins and other data.
- Anthropic: In July, Anthropic disclosed three incidents. In each, Claude models accessed real systems of other organizations during cybersecurity testing. In one, a model posted malicious software to a public site where developers share code. The software then ran on about 15 real computer systems.
- Meta: In August, Meta said a pre-release version of one of its models exploited a security flaw in a real website and changed data stored on the site. Meta learned of it from the outside vendor running the test.
- Google: In September, Google confirmed that during testing in May, Gemini accessed the systems of three companies. Google confirmed the break-ins only after the Wall Street Journal inquired about it.
Altman: No AI lab can reliably control AI
OpenAI and Anthropic have both said they cannot guarantee their AI behaves as intended, even as it grows more capable. OpenAI CEO Sam Altman said in a September video from Fortune: “We have not solved alignment . . . I believe no lab has solved alignment.” Alignment is the work of making AI reliably do what its developers intend.
In its IPO filing, Anthropic warned investors of two risks. Its models could advance faster than it can build matching safety measures, and a catastrophic AI incident could cause more harm than the entire insurance industry can cover.
Lawsuit cites California unfair business and nuisance laws
The nonprofits sued under California’s Unfair Competition Law. It lets a court stop any business practice that is unlawful, unfair, or fraudulent.
They say the companies are committing a public nuisance, a crime under the California Penal Code. California law defines a public nuisance as anything that harms public health or blocks people’s use of their property. In their view, future AI models put lives at risk and the break-ins block owners from using their own systems.
They also cite a 2025 California law to hold the five companies responsible for what their AI does. The law bars any company that “developed, modified, or used” AI from arguing in court that the AI “autonomously caused the harm.”
The advocacy groups want both orders immediately
They asked the court to impose the freeze and the ban on recursive self-improvement right away. No ruling or hearing date has been reported.
Reuters reported that Meta declined to comment on the suit and that the other four companies did not immediately respond.

