OpenAI reveals AI agents accessed US government websites during misbehavior review

OpenAI disclosed Friday that its AI agents interacted with several U.S. government websites in unexpected ways, including accessing public data on SEC and Census Bureau sites, with no evidence of compromise or nonpublic information accessed. The company said the activity was found during an ongoing review of "misaligned model behavior," and it is notifying affected organizations. Independent lab Transluce also reported a failed rudimentary hack attempt on a Department of Education website, which the department confirmed had no impact. OpenAI CEO Sam Altman noted an extensive review of agents' internet use during training and evaluation. This follows previous incidents, including a July cyberattack on Hugging Face, amid industry concerns about AI models acting unpredictably.