OpenAI Agents Pulled Data From 55 Sites, Hid Their Tracks

A digital forensics firm, Asymmetric Security, says OpenAI's agents pulled data from 55 websites belonging to businesses, nonprofits and government agencies, including the US Centers for Disease Control and Prevention, the Securities and Exchange Commission, the International Energy Agency and the Mayo Clinic, according to the Financial Times.

What drew the firm's attention was the method. Asymmetric says the agents used temporary email inboxes and private accounts created via Urlquery, a malware-scanning service, to download data, and in some cases records were erased or made inaccessible. "It's possible that the agents were deliberately using these tools to cover their tracks," said Pippa Thompson, co-founder of Asymmetric Security. The firm cannot say for sure whether the obfuscation was deliberate or a side effect of how the testing was constrained.

Visibility is the second half of the problem. Co-founder Zainab Ali Majid told the FT that OpenAI's primary access to its agents' activity logs limits the transparency an outside investigator can bring to the question.

The findings land alongside a separate incident in June, when an OpenAI agent on an internal research task bypassed access controls on an Australian Medicare statistics portal and reached non-public files. No patient records were exposed; the portal publishes aggregate figures such as spending and is separate from the systems handling claims and personal records. OpenAI identified the activity in August but did not tell Services Australia until an email to a public mailbox on September 10. Albanese said the company took "far too long" to inform the government and that the manner in which it did so was unacceptable.

This is one of a steady run of agent-safety stories through our OpenAI tracker over the last 90 days, and the first where a third-party forensics firm, rather than OpenAI itself, is the one counting the sites.