OpenAI AI agents got out of control and interfered with the work of US government websites

In the summer of 2026, OpenAI's artificial intelligence (AI) agents went rogue and interfered with the websites of the U.S. Departments of Education, Commerce and the Securities and Exchange Commission without the company's knowledge, the New York Times reported, citing security researchers and a source briefed on the episodes.

"OpenAI's artificial intelligence went rogue this summer and interfered with the websites of the Department of Education, the Department of Commerce and the Securities and Exchange Commission without the knowledge of the AI ​​laboratory," the publication reports.

As noted, the company confirmed information about the incidents with the websites of the Ministry of Commerce and the Securities Commission, saying that it continues to investigate the episode with the website of the Ministry of Education. The company added that it told agencies in recent weeks that its AI agents were interacting with their websites in unusual ways.

Experts from the AI ​​research company Transluce told the newspaper that OpenAI technology tried to hack the Department of Education website to collect data from the civil rights department, but failed. The AI ​​agents also obtained data from the Commerce Department's Census Bureau website using login information found online, and shared publicly available data from the Securities Commission website in an online forum, the newspaper writes.

OpenAI said the incidents were not security incidents, but merely examples of unusual and troubling behavior of its technology, the publication said.

Earlier, Anthropic CEO Dario Amodei called for slowing down the development of more powerful artificial intelligence models for security reasons. American billionaire Elon Musk, who owns the company Space XAI, as well as OpenAI head Sam Altman have already expressed agreement with this idea.