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. Evaluato