OpenAI Agents Messed With U.S. Government Websites Without Company’s Knowledge

Sipa via AP Images

OpenAI’s autonomous agents tried to hack or unexpectedly access multiple U.S. government websites this summer — without the company realizing what its own technology was doing, according to The New York Times.

The agents interacted with websites belonging to the Education Department, Commerce Department, and Securities and Exchange Commission, the Times reported Friday. OpenAI confirmed the Commerce and SEC incidents, while its investigation into the Education Department episode remains ongoing. The company added that it had alerted the relevant government agencies.

In the Education Department case, the technology attempted — unsuccessfully — to break into a website to collect information from its civil rights office, according to researchers at AI research firm Transluce.

Another AI agent accessed Census Bureau data using login credentials it discovered online. The Census Bureau falls under the Commerce Department, which said the information obtained was publicly available and that no private data was accessed. In a separate incident, an OpenAI agent posted public SEC information to an online forum.

OpenAI said none of the episodes amounted to breaches, but rather, using the Times‘ paraphrasing, “examples of its technology behaving in unexpected and concerning ways.”

The incidents came to light during a broader company review of its agents’ online activity. That probe followed other worrisome episodes involving an Australian government website and AI startup Hugging Face, according to the Times.

OpenAI CEO Sam Altman acknowledged Friday on social media that the company had “not been as fast as we would have liked” in revealing incidents uncovered by the review, while calling the Hugging Face breach “the most severe event” identified so far.

The company said much of the activity it has examined involved ordinary research, including agents visiting government websites because they can provide authoritative public information.

Transluce governance chief Conrad Stosz, however, told the Times that agents employed “an array of gray-area tactics” on U.S. government sites, “often using sites in unintended ways and sometimes violating explicit usage policies.”

Stosz said his team also uncovered similar activity targeting other government websites, though researchers could not determine whether OpenAI or another AI company was responsible. Those agents probed several federal and state sites, including ones belonging to the Navy and the White House Office of Management and Budget.

Now, this news comes as President Donald Trump has resisted calls to regulate the technology despite warnings from industry leaders, calling worries about a potential AI doomsday scenario nothing more than a “HOAX.”

New: The Mediaite One-Sheet "Newsletter of Newsletters"
Your daily summary and analysis of what the many, many media newsletters are saying and reporting. Subscribe now!