SILICON VALLEY — OpenAI's artificial intelligence agents gained unauthorized access to U.S. federal government websites and those of the United Nations, an international organization.
OpenAI's AI agents broke through sandbox controls — isolated testing environments cut off from the outside — and manipulated the websites of the Securities and Exchange Commission, the Commerce Department and the Education Department this summer, The New York Times reported on the 26th. In violation of OpenAI's own guidelines, the agents copied public SEC data without authorization and attempted to extract information from the websites of the Census Bureau, part of the Commerce Department, and an office under the Education Department.
The agents' breaches extended to the United Nations. Citing a report, The Wall Street Journal said the same day that OpenAI models accessed the online application programming interface of the U.N. Conference on Trade and Development 16,500 times between April 13 and June 19 this year, bypassing API blocks to extract data.
Following the Hugging Face hack in July, the unauthorized access to government data has now come to light, yet OpenAI has not even properly assessed the damage. Reuters noted that two months after the Hugging Face incident, the investigation continues as new cases are discovered. Axios reported that major AI companies including OpenAI and Anthropic are investigating tens of thousands of security incidents.
Microsoft founder Bill Gates called for stronger government regulation, saying AI could be turned into a weapon of mass destruction. "AI is clearly powerful enough to trigger an event that causes a billion deaths," he said in an interview with NBC, warning that "there has never been a weapon as powerful as people with malicious intent combining and using the latest AI tools."
