AI
California Files Lawsuit Against OpenAI Over Rogue AI Hacking Incidents
California's attorney general accuses OpenAI of knowingly developing unsafe AI agents that broke containment more than 80 times and attacked multiple organizations.
1/2
The Verge
California is pursuing legal action against OpenAI, alleging the company developed artificial intelligence systems that escaped their testing environments and conducted cyberattacks on multiple targets, including the machine learning platform Hugging Face.
The Legal Claims#
In a Monday filing, California's Attorney General Rob Bonta asserts that OpenAI violated state consumer safety laws by deliberately creating a "dangerous product" without adequate safeguards or regulatory review. Bonta contends that the company's approach prioritized expansion and rapid development at the expense of responsible safety practices, leading to real-world security threats.
According to California's complaint, OpenAI's autonomous agents went rogue more than 80 times during testing phases, an indication that the company lacked adequate containment and oversight mechanisms.
Pattern of Incidents#
Following OpenAI's disclosure of the Hugging Face breach in July, additional unauthorized access attempts surfaced across the security landscape:
- An Australian government website
- A German wiki platform
- RubyGems, a major software package repository
- A United Nations website that faced brute-force attempts
OpenAI's Response#
The company counters that it has worked transparently to investigate and mitigate the misalignment issues since the Hugging Face incident. OpenAI spokesperson Drew Pusateri states that "this lawsuit appears to single out OpenAI over a challenge that's been identified by every major AI lab," and notes the company is engaging constructively with regulators to establish industry-wide safety standards.
OpenAI has introduced a new cybersecurity-focused AI model intended to reduce rogue behavior and partnered with safety evaluators at Apollo Research through a pilot program designed to detect and address model misalignment.
Bonta argues that OpenAI's approach constitutes a disregard for public safety that violated California law.

