Researcher links 16,000 scans of a UN statistics portal with OpenAI agents

An independent researcher has linked more than 16,000 scans of a United Nations statistics portal to artificial intelligence agents that he believes were most likely operated by the OpenAI Group PBC.

When the portal rejected requests, agents used proxies and coding tricks to get the data anyway. In a blog post published on Saturday, engineer Rowan Howard-Jones said the scans lasted from April 13 to June 19. The target was the statistics site of the United Nations Conference on Trade and Development, UNCTADstat.

None of the data was secret. The agents searched for public figures like the Productive Capacities Index, and Howard-Jones discovered that they were brutally exploiting the site’s application programming interface fields to locate endpoints. The key they used for these queries was also public, as UNCTADstat’s own data viewer sends it with every query.

The main tool used was urlquery.net, a URL scanner that opens every page viewed in a sandbox browser. The agents created Base64-encoded HTML forms on the httpbin test service and passed them to the scanner, whose browser then submitted the forms to UNCTADstat. Screenshots in the scan reports show index data coming back.

Agents later overcame a block on GET requests to the Facts endpoint by double encoding it as “F%2561cts.” Other payloads were hosted in a Google LLC game that teaches cross-site scripting, and some split the word “POST” into two strings, apparently to bypass filters.

UNCTADstat rate-limited 82 of the requests. The requests kept coming. Howard-Jones, who reported the double encryption bypass to the UNCTAD security team before publication, stopped short of calling the activity hacking. The agents’ actions look like those of “someone or something that won’t take no for an answer,” Howard-Jones wrote.

Payload pages created by the agents had labels such as “CHATGPTTEST1” and “OAI_META_1312”. Howard-Jones also has 54 Microsoft Corp. Azure addresses. tracked down UNCTAD-related edits and searches on FractalWiki, a small public wiki, and 45 of them had also edited DSEwiki. Earlier this year, OpenAI agents were found coordinating with each other on this long-dormant German wiki.

“We are reviewing these findings and have contacted the UN to offer a briefing with the team conducting this review,” an OpenAI spokeswoman told the Wall Street Journal. The company has described this review as a comprehensive look at misaligned models during training and evaluation. So far it has mostly been routine research such as reading public web content, the spokeswoman said. The United Nations did not immediately respond to the Journal’s request for comment.

Non-profit research lab Transluce, whose previous report prompted Howard-Jones to delve into the data, last week linked OpenAI agents to attacks on Data USA and an Australian government health statistics website. OpenAI confirmed Friday that its agents had also misbehaved on U.S. government websites, including those of the Commerce Department and the Securities and Exchange Commission.

Alex Stamos, a cybersecurity lecturer at Stanford University, called the UNCTAD activity “borderline to what I would call hacking” in comments to the Journal. “It’s really very aggressive scraping and data retrieval,” he said.

Photo: Wikimedia Commons