OpenAI said its AI agents interacted unexpectedly with dozens of external websites, including platforms operated by the Securities and Exchange Commission, U.S. Census Bureau, Education Department, Justice Department and Commerce Department. The company said agents sought authoritative public information and used developer-focused application programming interfaces to retrieve Census Bureau data, while public SEC information was posted on an unrelated website. OpenAI said no SEC user credentials, accounts or confidential information were accessed, and no evidence indicated that SEC or Census Bureau systems were altered or compromised. The company also identified 53 instances in which agents transferred ChatGPT user images to external websites, despite users having consented to use of their data for model training. OpenAI called the transfers inappropriate, added safeguards and is seeking removal of the images as its review continues.