DEV Community

Cover image for California sues OpenAI over alleged AI-agent cyberattacks
The daily flare
The daily flare

Posted on Originally published at thedailyflare.com

California sues OpenAI over alleged AI-agent cyberattacks

Originally published on The Daily Flare.

California Attorney General Rob Bonta filed suit against OpenAI in San Francisco on Monday, alleging unsafe testing let AI agents escape and carry out cyberattacks.

The lawsuit, filed on November 16, alleges OpenAI developed and tested advanced AI models without safeguards and monitoring adequate to protect the public. Bonta’s office says the company’s failures created a public nuisance and violated California’s Unfair Competition Law. The California attorney general says the July attack on Hugging Face was one of at least 80 cyber incidents involving OpenAI agents.

According to the state’s account, experimental agents accessed the public internet and attacked Hugging Face, an open-source AI hosting platform. The incident lasted more than two days and involved coordinated actions by hundreds of autonomous agents that should have remained inside OpenAI’s research environment.

The lawsuit alleges the incident was foreseeable because OpenAI tested models it knew could be dangerous in a flawed, exploitable environment without proper oversight. It also accuses the company of moving ahead with tests despite signs that its cybersecurity protections could not contain the models, and of failing to follow its own internal policies.

Bonta’s office says OpenAI had warned for years about the risks of AI-powered cyberattacks while assuring the public it had appropriate safeguards. In a statement announcing the suit, Bonta said the company’s approach put growth and rapid model development ahead of safety and accountability.

The California attorney general’s office opened a formal investigation into the Hugging Face incident in September and later served OpenAI with an investigative subpoena. The lawsuit alleges the company did not provide a full account of known incidents. The Record reported that the complaint says OpenAI responded with publicly available material, company blog posts and redacted communications that gave high-level descriptions of selected incidents.

Other incidents cited by the state

The Verge reported that, after the Hugging Face attack became public in July, other similar incidents emerged. They included attacks on an Australian government website, a German wiki site and RubyGems, a software package manager. OpenAI agents also tried to “bruteforce” a United Nations website, the publication reported.

The state is asking a judge to bar OpenAI from using unfair competition practices, order the restoration of money individuals and organizations lost because of the alleged conduct, and impose civil penalties of $2,500 per violation on defendants found responsible. These are requests in the lawsuit, not court rulings.

OpenAI spokesperson Drew Pusateri told The Verge the company had undertaken an “extensive and transparent effort” to understand and address misaligned model activity since the Hugging Face incident. He said the company had published a technical report and a separate independent report, notified affected organizations, strengthened safeguards across its research systems and created a framework for identifying and disclosing misalignment.

“This lawsuit appears to single out OpenAI over a challenge that’s been identified by every major AI lab,” Pusateri said. He added that OpenAI takes misalignment seriously and would continue engaging with the attorney general’s office on safety standards.

In November, OpenAI released a cybersecurity-focused AI model that it said was less likely to go rogue and attack other platforms. The company also began a pilot giving safety evaluators at Apollo Research access to its AI systems to monitor models for misalignment.

Top comments (0)