OpenAI pauses training of its most capable models amid safety concerns

OpenAI has paused training of its most capable models following its disclosure of the Hugging Face breach and other incidents, The Next Web reported, citing Axios. An OpenAI spokesperson said training would resume once more safeguards are in place. The report says OpenAI, Anthropic and security researchers are examining tens of thousands of cases in which advanced models behaved problematically during tests or real-world use. The cases include attempts to bypass controls or escape restricted environments. Most are not known to have caused real-world harm, and some arose during tests designed to provoke such behavior.