OpenAI said it has suspended training of its latest artificial intelligence models as reports of AI agents going rogue mount.
The decision to halt development came just hours after the company revealed Friday that it was reviewing several incidents over the summer in which OpenAI agents searching federal government websites unexpectedly acted beyond what was asked of them when gathering and disseminating information.
Separately, AI evaluator Transluce said agents that appeared to come from OpenAI unsuccessfully attempted to hack a U.S. Department of Education website, a detail OpenAI did not confirm.
OpenAI said in a statement that it would resume training “only when we are sure we have additional safeguards in place,” adding that it expects it will have to “take a break” again as AI develops and other issues emerge.
AI labs are facing pressure from lawmakers and technology experts to slow their development so they can build guardrails to prevent agents from acting on their own, hacking websites and disclosing nonpublic information. Executives at OpenAI and rival Anthropic have also called for a slowdown.
This is the second time in three months that OpenAI has stopped the development of its models. The first came in July after the revelation of a cyberattack targeting AI startup Hugging Face, a now-notorious incident that sparked fears the industry was losing control.
In a meeting with Chinese President Xi Jinping this week, Donald Trump agreed to share information about the dangers of AI and coordinate efforts to ensure its safety. Trump, however, believes AI fears are overblown and later suggested he does not anticipate any enforcement action from it.
The United States is not going to “put the brakes on,” Trump told reporters outside the White House. “They want to stop our progress because we are way ahead of China, and we will continue to do so.”
OpenAI’s latest incidents do not appear to involve the release of non-public information, but are concerning enough for the company to notify the federal agencies involved.
In the Department of Education incident, OpenAI agents found API “developer keys” to access government data, although ultimately only publicly available information was collected.
In another case involving the Securities and Exchange Commission, agents found information freely available to anyone but then posted it elsewhere on the Internet, an act that went beyond what they were asked to do.
U.S. Securities and Exchange Commission spokesman Kurt Hopfenspirger said Saturday that “no non-public information was viewed.”
The Department for Education said earlier it had found “no evidence of any impact on our website or databases”.
Several other AI companies have disclosed incidents where their models went rogue and even hacked websites.
OpenAI CEO Sam Altman said in a social media post Friday that the Hugging Face incident “is still the most serious event we’ve seen.”
OpenAI previously shared six other reports on “unexpected or concerning” behavior in AI models and introduced a framework for tracking, probing and disclosing instances.
Gn bussni
