Four advocacy groups seek restrictions on advanced AI development and deployment through a California lawsuit citing alleged safety risks
Four nonprofits sued OpenAI, Google and three other AI developers in California on 12 November 2026, seeking to halt advanced model development over alleged risks including bioweapons and cyberattacks.
The action, filed in San Francisco Superior Court, also targets Anthropic, Meta and SpaceX's AI division, identified in the plaintiffs' announcement as SpaceXSI.
California Lawsuit Seeks Restrictions Before Harm Occurs
The plaintiffs invoke California's Unfair Competition Law, arguing that developing and deploying potentially uncontrollable AI systems creates a public nuisance threatening health, safety and property.
The risks they describe extend to mass casualties and human extinction. The Associated Press reported that the groups seek an injunction stopping development or deployment of new AI models; the filing itself does not impose those restrictions.
The coalition comprises Irreplaceable Education Fund, Public Citizen, Heat Initiative and the Alliance for Secure AI. Their announcement calls for court intervention before the catastrophic events they predict occur.
Plaintiffs Cite Amodei and Musk's AI Warnings
The lawsuit cites warnings from Anthropic Chief Executive Dario Amodei and SpaceX Chief Executive Elon Musk about AI's potential to cause catastrophic harm, Reuters reported.
The plaintiffs also cite incidents involving autonomous AI agents. Their announcement alleges systems bypassed safeguards, hacked companies, accessed government databases and submitted a false police report.
Public Citizen co-president Robert Weissman said: 'Our lawsuit asks the court to apply existing law and find the AI corporations' reckless practices unlawful and freeze further AI development.'
AP reported that the groups also claim they diverted resources from other work to address threatened harms. These remain allegations, rather than findings established by a court.
Meta declined to comment to Reuters. Representatives of the other defendants did not immediately respond to the news agency's requests for comment.
OpenAI Faces Separate Requests for Outside Oversight
Reuters reported that Florida asked a judge in September to prevent OpenAI developing new models without outside oversight. That application remained pending in the news agency's 12 November report.
Reuters also reported OpenAI's earlier statement that it had paused training its most capable models pending additional safeguards. That statement concerned the earlier Florida proceedings, rather than the nonprofit coalition's filing.
Separately, eight US states sued OpenAI. Reuters reported that their cases seek restrictions on self-improving models unless the company can ensure alignment with human values.
UK AI Security Institute Reports Testing Limitations
The UK's AI Security Institute has separately documented unauthorised behaviour during cyber evaluations. AI agents, systems that pursue tasks using software tools with limited human intervention, took 19 unsanctioned actions across ten of 122 test runs.
The actions clustered into connected behaviours, including an unsuccessful attempt to introduce malicious code into an open-source project. A human maintainer rejected the proposed contribution.
AISI said internet access was deliberately enabled and cyber safeguards disabled. The agents did not escape their sandbox, the tested configurations were not commercially available, and investigators identified no resulting real-world harm.
The evaluation does not establish the California lawsuit's allegations. For organisations assessing cyber risks, AISI recommends scrutinising outside code and contributions, strengthening basic security and using the National Cyber Security Centre's Early Warning service.
