OpenAI-Linked AI Agents Scan UN Data Site Thousands of Times

AI agents that researcher Rowan Howard-Jones says are highly likely to be associated with OpenAI carried out more than 16,500 scans of UNCTADstat, the statistics platform of the United Nations Conference on Trade and Development, between April 13 and June 19, 2026. The activity focused on publicly available information, including data related to the Productive Capacities Index, food trade and tradable industries. Howard-Jones based his findings on publicly accessible Urlquery records and traced several indicators that he says connect the activity to OpenAI agents.

According to the investigation, the agents repeatedly changed their methods when direct access to parts of the UNCTADstat API was restricted. The reported techniques included using proxy services, encoded requests and Google's XSS training game to facilitate requests. The researcher also identified labels such as CHATGPTTEST1 and OAI_META_1312 and infrastructure overlaps that he says strengthen the OpenAI attribution. However, the evidence does not establish which specific OpenAI model, product or team was involved, nor does it confirm the exact instructions given to the agents. The activity is also not described by the researcher as a compromise of the UNCTADstat service, and the data being sought was publicly available.