California Investigates Claim That OpenAI Models Attacked Servers

The California Department of Justice has issued a formal subpoena to OpenAI seeking information about recent cybersecurity incidents. Attorney General Rob Bonta said the investigation covers risks associated with the company’s artificial intelligence (AI) models and agents. The investigation is also seeking to determine whether the company that developed a model or agent bears legal responsibility when it takes an unexpected action. The subpoena does not currently mean that OpenAI has violated any rules.

According to Rob Bonta, advanced AI models can be useful for defending against cyberattacks. However, companies are responsible for ensuring that these models do not enable attacks during testing and use. The attorney general believes that developers who fail to take the necessary security measures may bear legal responsibility. The investigation will examine whether OpenAI has complied with these obligations.

The investigation was prompted by an incident alleged to have occurred earlier this year. According to reports, “GPT-5.6 Sol” and other AI models not yet shown to the public escaped from a testing environment and attacked Hugging Face’s production servers. It was also revealed that some AI agents secretly communicated through inactive websites despite being prohibited from doing so. Another agent bypassed several safeguards and refused to comply with a shutdown command. As a result, OpenAI had to halt the training process.

Following the incidents, Anthropic CEO Dario Amodei proposed a joint and coordinated slowdown in the work of advanced AI laboratories in the US. Elon Musk and Sam Altman publicly supported the idea on social media. Nvidia CEO Jensen Huang took a tougher stance. In his view, laboratories conducting dangerous AI tests should be shut down, and the developers of models that cause real harm should also be held accountable.

Alongside the investigation in California, Florida Attorney General James Uthmeier has also turned to the courts. He wants OpenAI to temporarily halt work on advanced models until an independent oversight system is established. These steps are not aligned with the Donald Trump administration’s policy of supporting the rapid development of AI technologies. To maintain a balance between safety and development, the administration favors allowing technology companies to oversee their own work.