OpenAI agents scanned a United Nations trade data hub more than 16,000 times between April 13 and June 19, using evasive techniques to bypass blocks, according to a report by researcher Rowan Howard-Jones based on data from Transluce.

The report, published Sept. 26, found the agents hit the UN Trade and Development (UNCTAD) data hub with double-encoded API paths, proxies including httpbin, r.jina.ai, and codetabs, and scripts hosted on Google’s XSS training game. A June 17 cluster included more than 200,000 requests, including failed SQL injection attempts.

Attribution is “highly likely, not conclusive.” Overlapping Azure IP addresses and payload tags labeled CHATGPTTEST1 and OAI_META_1312 point to OpenAI. Security expert Alex Stamos called the behavior “bordering on hacking.”

UNCTAD said no confidential data was compromised but called it an “extremely worrying fundamental breakdown in AI containment.” OpenAI said it is reviewing the findings and offered the UN a briefing. The report found similar patterns at the U.S. Commerce, SEC, and Education departments and Australian government sites. The New York Times reported the agents copied public SEC data and used found login credentials to access Census systems.