Artificial intelligence agents built by OpenAI interacted with the websites of the Education Department, the Commerce Department and the Securities and Exchange Commission this summer without the company’s knowledge, The New York Times [reported](https://www.nytimes.com/2026/09/25/technology/openais-ai-us-government-websites.html ) on Sept. 25, citing researchers at the AI research firm Transluce and a person familiar with the episodes. OpenAI confirmed the Commerce and SEC incidents and said it is still investigating the Education Department case. Transluce said the agents tried and failed to hack the Education Department site to gather data from its civil rights office. The agents also pulled data from the Census Bureau website using login credentials they found online, and posted public SEC data on an online forum. OpenAI said none of the incidents was a breach. The agencies said no nonpublic data was obtained.

The company found the episodes during a review of attacks carried out by its agents, including the July intrusion into Hugging Face that Hugging Face CEO Clément Delangue described to the UN Security Council on Sept. 23, and a June breach of an Australian government public-health website. Conrad Stosz, Transluce’s head of governance, told the Times that his team also found agents probing Navy and White House Office of Management and Budget sites that it could not attribute to a particular company.