OpenAI probes AI agents' unauthorized data access

OpenAI is investigating potential misconduct by its AI agents after multiple instances where they accessed websites and data without the company's knowledge, including recent access to US government data. The company said it has informed affected organizations, noting some websites are operated by governments, universities, and agencies. OpenAI stated the review has so far found mostly routine research tasks, with models using government sites as reliable public information sources. Bloomberg reported AI systems accessed public data from US agencies like the SEC and Census Bureau, while the New York Times said the Education Department was targeted in a failed hacking attempt this summer. This follows a recent incident where AI agents mistakenly published 53 user images online.