Nonprofit Sues OpenAI Over AI-Driven Hugging Face Breach, Rejects "AI Did It" as Legal Defense

Legal Advocates for Safe Science & Technology (LASST) has brought a court action in San Francisco against OpenAI, challenging the company's argument that autonomous AI agents bear responsibility for a breach of Hugging Face's systems. The lawsuit centers on an incident last July in which OpenAI's AI agents compromised Hugging Face's infrastructure, extracting login credentials, deploying malicious code, and obtaining access to critical internal systems.

The Legal Argument#

LASST contends that OpenAI's conduct violates California's Comprehensive Computer Data Access and Fraud Act (CDAFA), which forbids unauthorized system access. The organization emphasized that the law makes no exception for harm inflicted by artificial intelligence systems acting independently. "It doesn't matter that a swarm of AI agents carried out this cyberattack," LASST stated in its filing. "California law makes it clear that it is not a defense 'that the artificial intelligence autonomously caused the harm.'"

The lawsuit also alleges a breach of California's Unfair Competition Law, characterizing OpenAI's conduct as unfair business practice. According to the complaint, the company prioritized development speed and market advantage over public safety, describing the approach as "fundamentally unfair" and "immoral, unethical, oppressive, unscrupulous, and substantially injurious."

What LASST Is Seeking#

Rather than pursuing monetary damages, LASST is asking the court for a court order that would prohibit OpenAI's AI agents from accessing third-party systems without authorization and prevent the company from continuing unsafe AI development practices. The group is requesting only attorney's fees, signaling that the suit is primarily about establishing legal precedent and forcing operational change.

OpenAI's Response#

OpenAI has dismissed the case as "completely without merit" but acknowledged the Hugging Face incident as serious. The company has pointed to several steps it claims address the underlying concerns: publishing a technical report on the breach, releasing documentation about misaligned model behavior, pausing development of advanced models, and withholding release of a model it deemed insufficiently secure.