OpenAI says AI agents interacted with Education, Commerce, SEC websites in US

WASHINGTON (TNND) — OpenAI said Friday it has notified dozens of organizations after finding its AI agents, software designed to independently carry out multistep tasks online, may have bypassed security controls, disrupted services, or otherwise affected outside websites during training and testing, including several U.S. government sites .

The company said its agents interacted in unusual ways with websites run by the Education Department, Commerce Department, and Securities and Exchange Commission, according to security researchers and a person familiar with the incidents.

NEW YORK, NEW YORK - SEPTEMBER 23: CEO of OpenAI Sam Altman speaks during a United Nations (UN) Security Council meeting on artificial intelligence (AI) and international security under “Maintenance of International Peace and Security” during the 81st session of the United Nations (U.N.) General Assembly at U.N. Headquarters on September 23, 2026 in New York City. Tech industry leaders including the CEOs of OpenAI, Anthropic and Hugging Face briefed ambassadors on the risks of artificial intelligence. (Photo by Alexi J. Rosenfeld/Getty Images)

OpenAI confirmed incidents involving the Commerce Department and SEC and said it is still investigating activity involving the Education Department.

Researchers at AI firm Transluce said an OpenAI agent unsuccessfully tried to access data from the Education Department’s Office for Civil Rights website. The New York Times reported another agent used login credentials found online to access Census Bureau data from a Commerce Department website.

OpenAI agents also shared public SEC data online. The SEC said it is in with OpenAI and knows of no unsanctioned access to nonpublic information.

Commerce said the Census data was public and contained no private information, while Education found no impact to its website or databases. OpenAI said most incidents were low severity, with little or no evidence of meaningful impact; its review could take months.

OpenAI-Hugging-Face Incident-Technical-Report PDF preview View PDF

The disclosure follows a June incident in which an OpenAI agent gained unauthorized access to an Australian government health-statistics portal while researching public medical spending. Australian officials said the agent accessed aggregate health statistics and internal file names, but there was no evidence it accessed patient records or personal Medicare information.

OpenAI said it discovered the Australian incident Aug. 11 and notified the government Sept. 10. Australia publicly disclosed the incident this week and launched a task force to investigate.

The company said its models often turn to government websites as authoritative sources of public information while carrying out research tasks.

OAKLAND, CALIFORNIA - APRIL 27: Demonstrators protest against AI outside the courthouse at the Ronald V. Dellums Federal Building as jury selection begins in the lawsuit between Elon Musk and OpenAI on April 27, 2026 in Oakland, California. Elon Musk invested in OpenAI early on believing it would be a non-profit, but is now suing OpenAI and its CEO Sam Altman for allegedly deceiving him by developing OpenAI into a for-profit company. (Photo by Benjamin Fanjoy/Getty Images)