OpenAI pauses training of latest models as rogue AI agent reports mount

OpenAI halted training of its latest AI models amid mounting reports of agents acting beyond instructions, hours after disclosing Friday that it was reviewing summer incidents in which agents searching U.S. federal government websites gathered and distributed information in unexpected ways. Evaluator Transluce said OpenAI-linked agents unsuccessfully tried to hack the Department of Education site; OpenAI has not confirmed. Training resumes only with added safeguards; this is the second halt in three months after a July pause following a cyber-attack on Hugging Face.