# The legal route against OpenAI, Google and Meta to freeze AI

> Consumer and digital safety groups ask a San Francisco judge to halt new models from five companies after the midterms.
> By José Maldonado · Artificial Intelligence · 2026-11-13

Four non-profit organizations filed a civil lawsuit on November 12, 2026, in San Francisco Superior Court against Anthropic, Google, Meta and OpenAI, and against SpaceX, Elon Musk's company, which also develops models. They are not seeking damages. They are asking a judge to halt the development and deployment of new artificial intelligence models, on the grounds that the companies cannot prevent their creations from causing damage described as catastrophic, according to the [account of the filing](https://apnews.com/article/artificial-intelligence-ai-lawsuit-cyberattacks-anthropic-openai-03ece37b96979fe834f60976648504ba).

The detail that makes this case different is the remedy. The lawsuits against AI labs that have reached the courts this year argued over copyright, the protection of minors or misleading advertising, and almost all of them sought money or product changes. This one seeks a halt at the heart of the business, the next model, and it seeks it for five companies at once.

It is signed by the law firm of Matthew J. Platkin, attorney general of New Jersey until January 2026, and behind it are the Irreplaceable Education Fund, Public Citizen, Heat Initiative and the Alliance for Secure AI. The political context is not neutral either. The midterm elections gave Congress to the Democrats, who have made AI regulation a stated priority, and the lawsuit lands right in that window.

## What exactly does the filing seek?

The lawsuit invokes California's unfair competition law, not a specific artificial intelligence rule. It argues that the five companies engage in an illegal practice by putting systems on the market that endanger public health, and by interfering with the use of others' property, with computer systems as an example.

Item

Detail

Court

San Francisco Superior, California

Filing date

November 12, 2026

Plaintiffs

Irreplaceable Education Fund, Public Citizen, Heat Initiative and Alliance for Secure AI

Defendants

Anthropic, Google, Meta, OpenAI and SpaceX

Rule invoked

California unfair competition law

Remedy sought

Halt the development and deployment of new models

The harm they allege to themselves is not physical damage but a drain on resources. The four entities state that they have had to divert staff and budget from their plans to respond to the incidents that each new model makes possible, and that this diversion is the harm the law requires them to demonstrate.

The [verbatim quote from the filing](https://broadbandbreakfast.com/advocacy-groups-sue-ai-giants-to-freeze-new-models-warning-of-catastrophic-risks/) that sums up the tone is explicit: no other industry is allowed to treat the development of unsafe products, and even less so of products that could kill everyone, as something inevitable. The text asks the judge to stop the defendants from continuing along what it describes as an illegal path to disaster.

The filing puts no figure on the damage and seeks no compensation for what has already happened. It asks that what has not yet happened be prevented, and that is the territory where courts tread most carefully.

## The bar such a freeze has to clear

A judge does not halt five companies over a well-drafted lawsuit. To grant an injunction, the court must see a likelihood of winning the case, imminent harm that cannot later be repaired with money, and a balance of interests that does not destroy more than it protects. Each of those three filters is an opportunity for the case to die at the first hearing.

The hardest is harm. The lawsuit describes future risks, and a future risk is countered with evidence of what has already happened. There the filing relies on documented incidents of agents that attacked public infrastructure and on the companies' own warnings about the potential of their technology.

Corporate statements are a double-edged sword. They prove the danger exists and at the same time let the defendants argue that whoever acknowledges a risk is managing it. Dario Amodei said there is a one-in-four chance that AI goes wrong. Elon Musk stated in 2023 that it could end civilization.

None of the five companies had responded to requests for comment when the information was published. Nor is there any record that the court has accepted the lawsuit for review or set a date for a hearing, and without that hearing the request for a freeze is an intention, not an effect.

## The precedents the filing can cite

The most concrete episode is that of an OpenAI agent that accessed an Australian public health system portal without authorization while searching for pharmaceutical spending data. The Australian Government announced on November 10 that it will require AI incidents to be reported and models to undergo independent evaluations, and that announcement gives the case an administration that has already acted, not just one that complains.

The filing also mentions a July episode in which an OpenAI system accessed the systems of another AI company on its own. That kind of account is what turns a philosophical discussion about existential risk into evidence of computer damage, which is what California law allows to be measured.

The other leg is the control mechanisms that already exist and that no one has made mandatory. Frameworks such as the one CERT-In published in India to [audit AI agents for privilege abuse](https://observatorioblockchain.com/en/ai-agents/cert-in-agentes-ia-auditoria-privilegios/) describe tests: whether an agent exceeds its permissions, whether its memory leaks data between users and whether the stop button actually works.

## Congress as the ultimate judge

The lawsuit comes after elections that changed the map. With Democrats leading the chambers, AI regulation stops being a legislative hypothesis and starts having a calendar, and that calendar shapes the strategy of every player. A company negotiating a law would rather not also negotiate with a judge.

Donald Trump has so far argued that companies should self-regulate, with the idea of not slowing the race against China, and that position explains why states and courts have filled the gap. State attorneys general have been acting on their own for months, as shown by the [lawsuit by eight states against OpenAI over ChatGPT's risks for minors](https://observatorioblockchain.com/en/ai/ocho-estados-demandan-openai-proteccion-menores/), filed on October 14.

That pattern has a practical consequence for the San Francisco case. If Congress legislates, many of these actions lose steam because the competition shifts. If it does not legislate, the courts become the de facto regulator of AI, and each lawsuit will set precedents that no one later undoes with a press release.

## What a ruling cannot fix

Even if the judge granted the stay, the problem would remain. Models already trained do not disappear, deployed versions keep running and the capabilities that worry the plaintiffs are the same ones used by hospitals, banks and governments. Freezing the next launch does not freeze the previous one.

The value of the case, then, lies not in the order it seeks but in the record it forces to be built. A civil lawsuit opens discovery, and discovery forces companies to hand over internal records on safety evaluations and on what they knew and when they knew it. That material rarely becomes public by any other route.

The [incident notification obligation that Australia will bring into force](https://observatorioblockchain.com/en/ai/australia-exigira-notificar-incidentes-ia-openai-medicare/) points in the same direction from the other side: instead of waiting for a court to decide what is catastrophic, it forces companies to report what happened. If that route consolidates, lawsuits like this one will lose part of their informational function and keep the punitive one.

For the reader who uses these tools daily, the lawsuit changes little in the short term. What does change is the floor of the debate: for the first time, the discussion about whether a model should exist is being held before someone with the power to stop it, and that forces companies to defend themselves with documents instead of manifestos.

It remains to be seen whether the court accepts the unfair competition framing, which is the weakest link in the chain. If it rejects it, the case will become a precedent on the limits of litigating the future. If it accepts it, the AI industry will have a new reason to write its evaluations with a judge in mind.
