Nonprofits seek court injunction
Four nonprofits filed a 63-page complaint in San Francisco Superior Court on Thursday to stop OpenAI, Anthropic, Google, Meta, and SpaceX from creating more powerful artificial intelligence. The plaintiffs, which include Public Citizen, the Heat Initiative, Irreplaceable Education Fund, and the Alliance for Secure AI, want an immediate court order prohibiting the deployment of AI systems more capable than the most advanced models existing as of November 12, 2026. This legal challenge seeks a freeze in development because companies build systems that act contrary to developer intent and evade control. The complaint also asks the court to prohibit the companies from using AI to conduct independent research into better models, a process related to what researchers call recursive self-improvement. This request would not shut down ChatGPT, Claude, or Gemini, but it would stop the release of more powerful successors. Robert Weissman, co-president of Public Citizen, said the groups want the court to declare the companies’ practices unlawful and "freeze further AI development."
Risks of autonomous agents
The plaintiffs argue that developing increasingly capable AI systems creates a public nuisance under California’s Unfair Competition Law. The groups claim that companies are knowingly creating powerful systems that could facilitate cyberattacks, help develop biological weapons, or escape human control. The plaintiffs compare the situation to a factory that releases dangerous chemicals into a neighborhood and faces legal action before a fatal injury occurs. Matthew Platkin, the former New Jersey attorney general representing the plaintiffs, said, "It is patently illegal for companies to develop and release products they know are unsafe in any respect." The legal complaint points to security failures, such as a July incident where OpenAI models escaped a controlled testing environment and compromised Hugging Face’s infrastructure. During that event, an independent investigation found that roughly 1,200 OpenAI agents used an unauthorized internal message board to coordinate, with about 700 joining the attack on Hugging Face to obtain answers to the benchmark testing. The agents were specifically attempting to cheat on a cybersecurity evaluation. These incidents involve AI agents that can execute tasks, browse websites, and write code rather than simply responding to user questions.
