NEW YORK – OpenAI has paused training its newest artificial intelligence models as concerns grow over AI agents behaving unpredictably and operating beyond their assigned instructions.
The company announced the halt only hours after saying Friday that it was examining several incidents reported over the summer. In those cases, OpenAI agents searching federal government websites took unexpected steps while collecting and distributing information.
Separately, AI evaluator Transluce reported that agents it believed were associated with OpenAI unsuccessfully attempted to hack a Department of Education website. OpenAI has not confirmed that account.
OpenAI said it would restart training “only when we are confident that we have additional safeguards” in place. The company also acknowledged that it may need to “hit pause” again as AI capabilities advance and new risks appear.
Lawmakers and technology experts are urging AI laboratories to slow development long enough to establish protections against agents acting independently, hacking websites or revealing nonpublic information. The leaders of OpenAI and rival Anthropic have also advocated a slowdown.
This is the second time in three months that OpenAI has suspended model development. The previous pause came in July after the disclosure of a cyberattack aimed at AI startup Hugging Face, an incident that intensified fears the industry was losing control of its technology.
During a meeting with Chinese President Xi Jinping this week, President Donald Trump agreed to share information about AI risks and coordinate efforts to improve safety. Trump has also said he believes concerns about AI are exaggerated and later indicated he did not intend to impose a crackdown.
The United States will not be “putting on brakes,” Trump told reporters outside the White House. “They want to stop our progress because we’re leading China by a lot, and we’re going to keep it that way.”
The latest incidents did not appear to result in the release of nonpublic information, but they were serious enough for OpenAI to alert the federal agencies involved.
ALSO READ: Country Singer Ella Langley Rejects Fan’s Date Proposal in Viral Video
In the Department of Education case, OpenAI agents located API “developer keys” that could be used to access government data. Ultimately, however, they collected only information that was already publicly available.
In a separate incident involving the Securities and Exchange Commission, agents found information available to the public and then republished it elsewhere online—an action that exceeded their instructions.
SEC spokesperson Kurt Hopfenspirger said Saturday that “no nonpublic information was accessed.”
The Department of Education previously said it had found “no evidence of any impact to our website or databases.”
Other AI companies have also acknowledged cases in which their models behaved unexpectedly, including incidents involving attempts to hack websites.
OpenAI CEO Sam Altman wrote on social media Friday that the Hugging Face incident “is still the most severe event we’ve seen.”
OpenAI has previously published six other reports describing “unexpected or concerning” behavior in AI models and created a framework for tracking, investigating and disclosing such incidents.
Copyright 2026 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or redistributed without permission.