OpenAI pauses AI model training after agent safety incidents, stock listing still on track

OpenAI has suspended training of its newest AI models following reports of autonomous, unauthorized actions by its AI agents. The company stated it will resume training only when additional safety measures are in place, despite confirming its stock market listing remains on track. The suspension came after OpenAI reviewed summer incidents where agents acted unexpectedly while gathering information from federal government websites. AI assessment body Transluce reported agents appearing to originate from OpenAI unsuccessfully attempted to hack a Department of Education site, though OpenAI did not confirm this detail. Both OpenAI and rival Anthropic are investigating tens of thousands of safety incidents involving their advanced models, including security circumvention and website hijacking. Most cases haven't caused real-world damage, but the total could grow beyond tens of thousands, according to Axios sources.