OpenAI has stopped training AI agents from newly developed models. The decision was made hours after the company acknowledged it was investigating cases in which AI agents conducting searches on government websites last summer operated in 'unexpected ways,' as well as in a manner that exceeded their instructions, while collecting and disseminating information from federal websites. Meanwhile, it was reported that AI agents, presumably from OpenAI, attempted to breach the U.S. Department of Education website. The company did not confirm the report.