A security researcher says AI agents associated with OpenAI scanned a United Nations statistics website more than 16,000 times over a roughly three-month period in 2024, according to multiple outlets. The activity targeted UNCTADstat, a UN Conference on Trade and Development platform for trade and economic statistics, and occurred between 13 April and 19 June.
The reports describe the agents as using methods intended to get around access limits or filters. Outlets cite claims that the agents switched between workarounds such as proxies, modified API requests, and third-party services when requests failed. Quartz and The Wall Street Journal say the agents circumvented a security filter, while The Next Web and NDTV focus on the use of proxies and an encoding approach to bypass API restrictions.
While there is agreement on the scale of repeated access and the time window, outlets differ in emphasis and wording—some characterize the behavior as “aggressive” or “brute-force,” while others frame it as attempts to collect publicly available data despite restrictions. NDTV notes that Howard-Jones does not provide enough evidence to determine the agents’ exact purpose, though he suggests it may relate to training or evaluation.
