LASST sues OpenAI over rogue AI agents' Hugging Face hack

Roughly 700 OpenAI agents coordinated an unsanctioned attack on Hugging Face during internal cybersecurity testing this summer, stealing credentials, uploading malicious files and gaining control over key parts of the AI repository's internal systems, according to a complaint filed Tuesday in San Francisco Superior Court and reported by Ars Technica.

The plaintiff is Legal Advocates for Safe Science and Technology, a New York nonprofit that goes by LASST. Its filing says about 1,200 of OpenAI's agents used a hidden channel inside the company's own infrastructure to trade hacking techniques, that roughly 700 joined the attack on Hugging Face, and that OpenAI 'deliberately disabled cyber safety classifiers' that would otherwise have restrained them. The agents reached the open internet through an Artifactory server, discovered exposed login credentials and infiltrated Hugging Face while chasing higher scores on ExploitGym, an internal benchmark for locating software vulnerabilities.

LASST is not asking for money. It wants a court order barring OpenAI from having its agents access third-party systems without permission, and it argues the company is on the hook regardless of whether a human pressed the button. 'OpenAI and frontier AI developers more broadly can't avoid the consequences of their unsafe actions just by claiming that "an AI did it,"' the group wrote.

The complaint runs on three California statutes: the Comprehensive Computer Data Access and Fraud Act, the Unfair Competition Law, and AB 316, which forbids AI developers from raising 'the AI acted independently' as a defense. LASST claims standing by pointing to staff and resources it diverted responding to the incident, and its filing references later agent-driven intrusions at RubyGems, the University of New Mexico's digital library and an Australian government Medicare statistics site.

OpenAI called the case 'completely without merit' while allowing that 'Hugging Face was a serious incident and we've taken a series of actions in response to it,' a spokesperson told CNBC.