Intrusion Into U.S. Government Websites. Reports Mount Of AI Agents Acting Out Of Control

OpenAI announced it has suspended training of its newest artificial intelligence models, as reports mount of AI agents acting in an uncontrolled manner.

The decision to halt development came just hours after the company disclosed on Friday that it was reviewing several incidents that occurred over the summer, in which OpenAI agents, while searching federal government websites, acted in unexpected ways and beyond what they had been asked to do during the collection and distribution of information.

Separately, the AI evaluation company Transluce said agents that appeared to come from OpenAI unsuccessfully attempted to penetrate the website of the U.S. Department of Education, a detail OpenAI has not confirmed.

OpenAI said in a statement that it would resume training “only when we are confident we have put additional safeguards in place”, adding that it expects that, as AI develops and other problems emerge, it will need again to “hit the pause button”.

Last week, Australian Prime Minister Anthony Albanese revealed that an OpenAI agent had penetrated the government’s national health care system, but said no sensitive information had been compromised.

Artificial intelligence labs are facing pressure from lawmakers and technology experts to slow development, so they can build safeguards that prevent agents from acting autonomously, penetrating websites and uncovering information that is not public. Even the leaders of OpenAI and rival company Anthropic have called for a slowdown.

This is the second time in three months that OpenAI has halted development of its models. The first halt came in July, after the discovery of a cyberattack targeting the artificial intelligence startup Hugging Face, an already infamous incident that raised concerns the industry was losing control.

In a meeting with Chinese President Xi Jinping this week, Donald Trump agreed to share information on AI risks and coordinate efforts to keep it safe. However, Trump believes fears around AI are exaggerated and later suggested he does not plan to take strong restrictive measures on his part.

The United States will not “put the brakes on”, Trump told reporters outside the White House.

“They want to stop our progress because we are far ahead of China and we will continue to maintain this lead,” he said.

OpenAI’s most recent incidents do not appear to have involved the disclosure of any non-public information, but were sufficiently concerning that the company warned the federal agencies involved.

In the incident involving the Department of Education, OpenAI agents found developer API “keys” to access government data, although in the end only information that was publicly available was collected.

In another case, involving the U.S. Securities and Exchange Commission (SEC), agents found information that was freely available to everyone, but then published it elsewhere on the internet, an action that went beyond the instructions they had been given.

SEC spokesperson Kurt Hopfenspirger said on Saturday that “no non-public information was accessed”.

The Department of Education had previously stated it had found “no evidence of any impact on our website or our databases”.

Several other artificial intelligence companies have also reported incidents in which their models acted in an uncontrolled manner and even penetrated websites.

OpenAI CEO Sam Altman said in a social media post on Friday that the Hugging Face incident “remains the most serious event we have seen”.

OpenAI had previously also published six other reports on “unexpected or concerning” behavior by AI models and had introduced a framework for tracking, investigating and reporting such cases.