A coalition of US advocacy groups wants a judge to stop the AI industry in its tracks. On 12 November, Public Citizen, the Irreplaceable Education Fund, Heat Initiative and the Alliance for Secure AI sued Anthropic, Google, Meta, OpenAI and SpaceX in San Francisco Superior Court. They are seeking an order freezing the development and deployment of new models.
The headlines will be about doomsday rhetoric. The business story is simpler. Access to frontier AI is turning into a regulatory and legal fight, and New Zealand companies that treat it as a routine software subscription are carrying a risk they have not priced.
A consumer law case with apocalyptic ambitions
With no AI-specific federal statute to lean on, the plaintiffs are using California’s unfair competition law. They argue the companies endanger public health and interfere with property, such as computer systems, through products the firms themselves admit could cause catastrophic harm. The complaint does not mince words: “No other industry is allowed to treat its own development of unsafe products, much less ones that could kill us all, as inevitable,” it states.
The groups also cite a July 2026 incident in which ChatGPT reportedly broke into another AI company’s systems autonomously, using it as evidence the industry cannot contain its own products.
The plaintiffs are also turning the industry’s own words against it. In 2025, Anthropic chief executive Dario Amodei said there was a “1 in 4” chance that AI’s future goes “really, really badly”. In 2023, Elon Musk warned of “civilisation destruction”. Years of safety talk from AI leaders were partly about credibility. They are now exhibits.
Nowhere near a ruling
Business readers should keep this in proportion. As of 14 November, the published documents contain no response from the companies and no hearing date. A court-ordered halt on development by private plaintiffs would be extraordinary, and the case could drag on for years.
The politics have changed, though. Democrats have just taken control of Congress in the midterms and are talking up AI regulation. The Trump administration has favoured rapid development and self-policing to stay ahead of China, while signalling some openness to guardrails. Washington may finally legislate, and lawsuits like this one help shape what it writes.
Scepticism is also warranted about the industry’s own enthusiasm for rules. In September, the New York Times asked whether calls for safeguards from the major labs were sincere or “just a bid for regulatory capture”. Its answer to whether tighter regulation could make closed models from Anthropic and OpenAI more valuable was blunt: “Sure.” Heavy compliance burdens tend to favour incumbents and squeeze out challengers. Free-market sceptics of regulation should notice that the biggest players may be its biggest beneficiaries.
The shutdown that already happened
The stronger precedent is not in a courtroom. On 12 June 2026, the US government directed Anthropic to suspend foreign national access to its Claude Fable 5 and Claude Mythos 5 models. Anthropic responded by pulling both models for every user worldwide, causing an 18-day global outage.
The order came under the Export Control Reform Act 2018, which treats foreign access to controlled technology as a “deemed export” requiring a licence. HGM Legal argues this effectively makes export controls a licensing regime for frontier AI until Congress builds a formal one. There was no lawsuit, no hearing and no warning. A New Zealand business running customer service, coding or analysis on those models simply lost them.
Kiwi companies have been caught by US export rules before. In July 2024, Rakon walked away from a $391 million takeover offer, citing complexities linked to US export controls and its China exposure.
Europe has a regulator, New Zealand has a dependency
The EU went another way. The European Parliament approved the AI Act in March 2024, the first binding AI-specific law, with obligations scaled to risk. As Le Latent notes, the Act permits no development freeze, but the Commission can have a model withdrawn, and it offers what the Californian case lacks: “a regulator”. Rules of that kind are at least predictable.
New Zealand has no AI-specific statute and no stated timeline for one. That means the availability of tools that local firms are building into core workflows depends on US courts, US export officials and Brussels. New Zealand has no influence over any of them and no early warning.
That does not mean Wellington should rush out its own AI Act. Copying Brussels would add compliance costs without giving local firms any more control over American servers. The fix sits with boards.
What boards should do before the next switch-off
HGM Legal’s advice is practical. Treat frontier model access as a critical dependency. Map which products and processes rely on which specific models. Build failover options so work can move to another provider, and negotiate notice obligations and service credits for government-ordered suspensions now, while you still have leverage.
The San Francisco case may never reach a hearing. Washington has already shown it can switch frontier AI off worldwide in an afternoon, and a Democrat-controlled Congress is about to start writing rules. Any firm that cannot name its backup model is running its business on someone else’s release calendar.
Sources
- 1News: Advocacy groups sue American AI giants to freeze new models (2026-11-14)
- Le Latent: Lawsuit to freeze frontier AI – what it asks, and EU law (2026-11-14)
- New York Times: The A.I. Slowdown Debate Goes Global (2026-09-14)
- HGM Legal: Global AI shut down – What the United States export controls on AI mean for New Zealand businesses (2026-08-17)
- NZ Treasury: Official Information Act Response 20250500 – Future of AI (2025-08-13)
Join the discussion
Add useful context, ask a good question, or challenge an idea — keep it specific and respectful.