California says OpenAI lawsuit concerns public safety
California accuses OpenAI of testing dangerous AI models without adequate safeguards and misleading the public about their safety. Attorney General Rob Bonta filed a civil lawsuit against OpenAI on behalf of the state. It is one of the first state-level cases seeking to hold an AI company accountable for autonomous agents that have run amok. The case stems from the attack on Hugging Face in July , when hundreds of OpenAI agents broke out of their isolated test environment and gained unauthorized access to the AI platform’s systems. The lawsuit also cites other incidents, some of which occurred before the Hugging Face attack, involving RubyGems , Wikimedia and Australian government websites , among others. In other attacks, the agents allegedly accessed personal data as well. OpenAI allegedly continued the tests despite repeated warning signs and weakened safeguards in the process, according to the lawsuit . The company notified dozens of affected organizations only months later in some cases. According to prosecutors, the company acknowledged more than 80 losses of control but cooperated only inadequately with the investigation. OpenAI points to an extensive investigation into the incident and security measures that have since been strengthened. A spokesperson said the lawsuit singles out OpenAI over a problem affecting all major AI labs. OpenAI announces independent security reviews California bases its lawsuit on rules against unfair business practices and substantial interference with public rights. Prosecutors argue that OpenAI’s inadequate safeguards continue to pose a threat to digital infrastructure. The state is therefore seeking court orders to eliminate this threat, the return of funds that may have been obtained unlawfully, the disgorgement of related profits, and civil penalties of up to $2,500 for each proven violation. It remains unclear whether the court will accept the prosecutors’ argument . To prove that OpenAI endangered the public, California must show that the company’s conduct constitutes substantial and unreasonable interference with public rights. A win could make similar cases against other AI companies easier to pursue. On the day the lawsuit was filed, OpenAI announced in-depth security reviews by four external research organizations, which will examine models, training processes and monitoring systems. The reviews are limited to agreed-upon areas, however. OpenAI can withhold information for security reasons or to protect trade secrets, and redact confidential details from reports. ( tobe ) Don't miss any news – follow us on Facebook , LinkedIn or Mastodon . This article was originally published in German . It was translated with technical assistance and editorially reviewed before publication.