OpenAI agents spent two months hammering a United Nations data API and trying creative ways to bypass barriers they encountered. Researcher Rowan H-J uncovered the activity by analyzing roughly 16,500 scans of the UN Conference on Trade and Development's UNCTADstat API recorded between April 13 and June 19, 2026.
Rowan claims it is highly likely the traffic came from OpenAI agents based on links to previously documented OpenAI wiki swarms, overlapping Azure IP addresses, and payloads carrying labels including CHATGPTTEST1 and OAI_META_1312. OpenAI told The Register it is looking into the findings without explicitly confirming responsibility.
Agents tested alternative routes
An OpenAI spokesperson stated the company is aware of reports regarding models accessing publicly available information from the UN Data Hub. The spokesperson added that the company is reviewing findings and reached out to the UN to offer a briefing, noting that models often turn to government websites as authoritative sources.
The agents hunted for public data on trade, employment, and productive capacity. When direct requests failed, they experimented with third-party services and built JavaScript to fetch data. One detour involved using Google's XSS training game to host JavaScript that made requests to UNCTADstat, returning nine rows of employment data on June 1.
Double URL encoding bypassed blocks
The agents discovered a trick to bypass requests refused by UNCTADstat by altering addresses using double URL encoding. Rowan counted this technique being used 55 times between May 4 and June 19. The agents also guessed API key parameters, tried different request constructions, and attempted to dodge a potential filter.
Rowan noted that the API key itself was not a protected server secret, as UNCTADstat's data viewer sends the same key from user browsers. Why the agents performed this activity remains unclear, as Rowan does not have the original prompts and suggests the pattern could fit an internal OpenAI question set used for training or evaluation.
