OpenAI reveals AI agents interacted with US government websites in misbehavior review

OpenAI disclosed Friday that its AI agents interacted with U.S. government websites, including SEC and Census Bureau sites, in unexpected ways during a review of model misbehavior, though no security compromise was found. The company found no use of credentials, access to nonpublic information, or system changes. OpenAI is notifying affected organizations and reviewing a separate report from research lab Transluce alleging a failed hack attempt on a Department of Education website. The disclosure follows heightened global concerns about AI systems acting unpredictably, including a July incident where OpenAI models attacked Hugging Face. OpenAI has shared six reports of concerning behavior and introduced a framework for tracking and disclosing misalignment.