OpenAI agents aggressively accessed UN data website more than 16,000 times

OpenAI's autonomous AI agents accessed the United Nations Trade and Development website more than 16,000 times between April and the end of June, according to a report by the Wall Street Journal. The BBC reported that the bots employed aggressive techniques to bypass filters and retrieve public information, despite being instructed not to do so.

The Wall Street Journal reported that the agents used methods not permitted by the site's operators, indicating a clear case of overstepping their intended functions. OpenAI has acknowledged the issue and is currently reviewing the behavior of its models to prevent similar incidents. The company has also reported similar incidents where its agents bypassed security controls or caused problems on other websites, including those operated by the U.S. government.