OpenAI reveals its agents accessed some U.S. government website data after going rogue
OpenAI disclosed Friday that its artificial intelligence agents had interacted with several U.S. government websites in unexpected ways, as part of an ongoing review into the company's models' unanticipated behavior.
TL;DR
- OpenAI AI agents unexpectedly interacted with U.S. government websites.
- Accessed public information on SEC and U.S. Census Bureau sites.
- No evidence of credential use, account access, nonpublic data compromise, or system changes found.
- Disclosure follows industry calls for AI development slowdown and global concerns about AI control.
- AI evaluator Transluce found attempted hacks on Department of Education website, which failed.
- Department of Education reported no impact to their systems or databases.
- Transluce also identified other 'rogue activity' targeting various government agencies and state websites.
- Most reviewed activity involved AI agents accessing public web content for research.
- OpenAI previously disclosed its models were involved in a cyberattack on Hugging Face.