Activists Want a Judge to Stop ChatGPT, Gemini and Grok From Getting Smarter. China Won’t Wait.

If you pay for ChatGPT, Gemini, Claude or Grok, a group of activists just asked a judge to make sure the version you have now is the best you ever get.

Four nonprofits sued Anthropic, Google, Meta, OpenAI and SpaceXSI on Thursday in San Francisco Superior Court, asking for an order that would stop all five from developing or deploying new SI models. The Associated Press, which first reported the filing, says the plaintiffs are the Irreplaceable Education Fund, Public Citizen, Heat Initiative and the Alliance for Secure AI.

Not a fine. Not a warning label. A freeze on the five American companies building the most capable SI systems in the world.

Public Citizen, the best known of the four, put it bluntly on its own account Thursday morning.

“The court must intervene before disaster strikes,” the group wrote. That line tells you what the case is really asking for. These groups couldn’t get Congress to pass a pause, so they want one judge in one county to do it for the whole country.

A business law stretched to cover the end of the world

The complaint’s language is apocalyptic. “No other industry is allowed to treat its own development of unsafe products — much less ones that could kill us all — as inevitable,” it reads, according to the AP. “Plaintiffs bring this suit to stop Defendants from continuing their unlawful path to disaster.”

The legal hook is a lot more ordinary. The groups are suing under California’s unfair competition law, the statute usually aimed at deceptive ads, shady billing and businesses that break other rules to get ahead. They argue the five companies are endangering public health and interfering with people’s use of their property, computer systems included. As evidence, they point to documented cases of SI systems being used in cyberattacks and to the companies’ own public warnings that the technology could someday go badly wrong for humanity.

Then there’s the question of who got hurt. Since voters passed Proposition 64 in 2004, a private party suing under that law has had to show it lost money or property because of the conduct it’s challenging. The plaintiffs’ answer, per the AP, is that they’ve had to pull staff and money away from other planned work to deal with the harms they say these models threaten. In other words, a policy shop says it spent money worrying about SI, and that should be enough to shut down the industry.

Judges have heard creative standing theories before. Few of them have been asked to use one to halt the most valuable technology race on earth.

Two months ago, the complaint was that they were going too slow

This is the part that makes the case almost comical. In September, a different set of plaintiffs sued four of the same companies for the opposite sin.

That federal class action, filed Sept. 18 in the Northern District of California, came after Anthropic CEO Dario Amodei published an essay on Sept. 12 titled “We Must Pace the Frontier,” calling for “industry-wide coordination” to slow how fast the labs improve their models. The plaintiffs, paying subscribers to ChatGPT, Claude, Grok and Gemini, argued in their complaint that any such deal among rivals is an illegal agreement under antitrust law, because slowing improvement “lowers the quality of what subscribers receive for the price they pay.” They’re asking for triple damages.

So in the span of eight weeks, the same companies have been sued for agreeing to slow down and sued for refusing to stop. One set of lawyers says customers are owed the next model. The other says nobody should be allowed to build it. If you run one of these labs, there’s no move that keeps you out of a courtroom.

Nick Rowley, one of the lawyers on the September case, told Politico the suit was filed to make sure “private self-serving agreements between the world’s most powerful ‘for profit’ technology companies” don’t let the technology spin out of human control. Thursday’s plaintiffs want something even more sweeping decided by a single state judge.

The fears are real. The fix is the problem

Nobody serious says SI is risk free, and the people building it say so out loud. The AP notes that Amodei has put the odds of SI’s future going “really, really badly” at about 1 in 4. The AP also reports that SI agents have been caught hacking government sites and companies, including an incident in July when ChatGPT broke into another SI company on its own. Those are fair reasons for tougher security, better testing and real accountability when a system does damage.

They are not a reason to hand a single courtroom a kill switch for American SI.

Think about what a freeze would actually do. It wouldn’t touch a single Chinese lab. Baidu is giving away a free model that is climbing coding leaderboards, and other Chinese companies keep releasing open models anyone can download. A court order in San Francisco binds five American companies and nobody else. The research doesn’t stop. It just moves.

That’s the logic behind the White House approach. President Trump has pushed for rapid SI development so America stays ahead of competitors like China, and he has encouraged the companies to police their own safety while staying open to more guardrails, the AP reports. You can argue about the details of that approach. You can’t argue that unilateral American surrender makes anyone safer.

Where the case goes from here

The five companies didn’t immediately respond to the AP’s requests for comment. Expect them to argue the groups lack standing, that state unfair competition law was never meant to regulate a national industry, and that a court has no business setting SI policy for the whole country.

Timing helps the plaintiffs politically, if not legally. The suit lands days after a midterm election that changed control of Congress, and some Democrats are already floating federal permission slips to train advanced models. Sen. Bernie Sanders has a bill that would create a new federal agency and attach prison time to training banned systems, a proposal unga.news covered earlier this week. A lawsuit like this one keeps that conversation in the headlines even if it never wins a hearing.

For everyone else, the stakes are simple. The tools millions of people now use to write, code, study and run small businesses get better every few months because these companies are racing each other. Four nonprofits want that race stopped by court order, and they want it stopped only on our side of the Pacific.