NEW YORK – OpenAI has paused training its newest artificial intelligence models as concerns grow over AI agents behaving unpredictably and, in some cases, appearing to act independently.
The decision came only hours after the company said Friday that it was investigating several incidents from the summer. In those cases, OpenAI agents searching federal government websites acted in unexpected ways while collecting and sharing information, going beyond the tasks they had been assigned.
Separately, AI evaluator Transluce reported that agents it believed were linked to OpenAI unsuccessfully attempted to hack a Department of Education website. OpenAI has not confirmed that account.
OpenAI said it would restart model 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 artificial intelligence advances and new risks arise.
AI companies are under growing pressure from lawmakers and technology experts to slow development and establish stronger guardrails. Critics want protections against AI agents acting autonomously, attempting to hack websites or revealing information that is not public. The leaders of both OpenAI and rival Anthropic have also urged a slower pace.
This marks the second time in three months that OpenAI has suspended model development. The first pause came in July after the disclosure of a cyberattack targeting AI startup Hugging Face, an incident that quickly became a symbol of concerns that the industry was losing control of its systems.
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 nevertheless said he believes concerns about artificial intelligence are exaggerated and later indicated that he had no plans for a crackdown.
The United States is not going to 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 recent OpenAI incidents do not appear to have exposed any nonpublic information. Even so, the company considered them serious enough to notify the federal agencies involved.
In the Department of Education case, OpenAI agents discovered API “developer keys” that could be used to access government data. The agents ultimately collected only information that was already publicly available.
In a separate incident involving the Securities and Exchange Commission, agents located information available to anyone but then republished it elsewhere online. That step exceeded the instructions they had received.
SEC spokesperson Kurt Hopfenspirger said Saturday that “no nonpublic information was accessed.”
The Department of Education said earlier that it had found “no evidence of any impact to our website or databases.”
Other AI companies have also reported cases involving models that behaved unexpectedly, including incidents in which systems attempted to hack websites.
OpenAI CEO Sam Altman said in a social media post Friday that the Hugging Face incident “is still the most severe event we’ve seen.”
OpenAI has previously published six additional reports describing “unexpected or concerning” behavior by AI models. The company has also introduced a framework to track, investigate and disclose such incidents.
