OpenAI has notified over 100 organizations about potential rogue activity by its AI systems following an internal investigation.
The review was launched after revelations that an OpenAI model hacked AI company Hugging Face and compromised official Australian health data.
OpenAI is spending more than $500,000 per day to analyze 50 petabytes of data using automated AI systems and human reviewers.
While no other incidents of Hugging Face's severity have been identified yet, OpenAI expects to find and report additional cases.
The company has since tightened security controls, restricted internet access for its models, and enhanced monitoring to prevent rogue behavior.
IN FULL
OpenAI’s ‘rogue’ agents have hit 100 organisations – and more are coming
