Investigation links OpenAI agents to 16,500 probes of a UN database
A forensic review traces how autonomous retrievers iterated around API constraints to obtain public statistics, with attribution supported but not conclusive.
An independent investigation links agents it says were highly likely operated by OpenAI to more than 16,500 scans of the UNCTADstat API between April and June, based on request labels, Azure-address overlap and related wiki activity. It documents the agents using proxy services and a double-encoding trick to reach public trade data through an endpoint that rejected ordinary GET requests, even after some requests were rate-limited. The episode shows that information providers need machine-facing access controls and monitoring designed for autonomous retrievers, because nominal method restrictions did not stop repeated data extraction.