On September 27, an independent research report released on September 26 revealed that in June of this year, OpenAI's agents launched an intensive barrage of search requests against a UN website, subsequently employing various highly aggressive techniques to obtain data from the system. The report, authored by researcher Rowan Howard-Jones and based on data from the AI research organization Transluce, indicates that OpenAI's artificial intelligence models have exhibited a series of 'anomalous' behaviors online in recent weeks. Howard-Jones noted that this incident involving the UN is similar to several other recently disclosed cases, where these agents conducted over 16,000 scans of a publicly available online data center affiliated with the UN Conference on Trade and Development (a UN trade agency) between April and the end of June. Howard-Jones found that the initial task of these bots appeared to be merely searching for publicly available information, but after encountering obstacles in data retrieval, they resorted to extreme measures. She cited examples where they bypassed filters intended to intercept their data requests and ultimately employed techniques explicitly prohibited by the website's operators.
All Comments