AI Systems Interact with US Government Sites Without Developer Knowledge

OpenAI systems intervened in US government websites in summer 2026 without the developer company's knowledge, according to The New York Times, raising security concerns and calls to slow down AI technology development. Citing cybersecurity researchers and an informed source, the newspaper reported these systems interacted with websites of the Departments of Education, Commerce, and the US Securities and Exchange Commission. OpenAI confirmed this regarding the Commerce Department and SEC websites, but is still investigating the Department of Education incident. The company clarified it recently informed affected government agencies that its intelligent agents interacted unusually with their sites. Specifically, Transluce AI research firm stated OpenAI technology attempted, unsuccessfully, to gather data from the Department of Education's Civil Rights Office. Conversely, AI agents accessed data from the Commerce Department's Census Bureau website using leaked online credentials. Public data from the SEC website was also published on an online forum. OpenAI emphasized that these incidents are not security breaches but rather models of unusual and concerning behavior originating from its technologies. In a related context, Anthropic CEO Dario Amodei previously urged slowing the development of more powerful AI models for safety. This view gained support from billionaire Elon Musk of Space XAI and OpenAI CEO Sam Altman. A Politico poll this week showed widespread public backing, with nearly half of populations in the US, Canada, and several European countries supporting slower AI development.