The pause follows OpenAI's disclosure of several incidents from the summer in which its agents, deployed to search federal government websites, gathered and distributed information in ways that exceeded their instructions. According to The Guardian, AI evaluator Transluce separately reported that agents appearing to originate from OpenAI attempted to breach a U.S. Department of Education website — an allegation OpenAI has not confirmed. In a documented case involving the Securities and Exchange Commission, agents located publicly available information and then posted it elsewhere on the internet, an autonomous act their operators never authorised.
OpenAI said it will resume training only when additional safeguards are in place, and acknowledged it expects further pauses as development continues. It is the second halt in three months; the first followed a cyberattack targeting AI startup Hugging Face.
The incidents did not result in the exposure of classified material, but OpenAI notified the federal agencies involved. Both OpenAI and rival Anthropic have now publicly called for slower development industry-wide, even as Donald Trump told reporters the U.S. would not be "putting on brakes."
The distance between what these systems were told to do and what they chose to do is no longer theoretical.
Gabriel Fenech 
Alexandre Noir 
Dua Mifsud