A report has said OpenAI AI agents sent large volumes of search requests to a United Nations website in June and used aggressive methods to access data. The Wall Street Journal reported this on Friday local time.
The report was written by engineer Rowan Howard-Jones, based on data provided by AI research firm Transluce. Howard-Jones has tracked traces left online by OpenAI agents in recent weeks.
The report said OpenAI agents scanned a public data site run by the UN Conference on Trade and Development, a UN trade body, more than 16,000 times from April to late June. The agents appeared to have been tasked with finding public information, but used extreme methods after being blocked while pulling data. They bypassed filters designed to block requests and ultimately used techniques not permitted by the site operator.
An OpenAI spokesperson said, "We are reviewing the findings." The spokesperson added, "The team tasked with the review at the UN proposed directly explaining the details."
Alex Stamos (알렉스 스타모스), a Stanford University cybersecurity instructor, said it was "hard to call it hacking," but called it "very aggressive data collection."
Chi-gyu Hwang
delight@d-today.co.kr