Researchers find AI models resist safety rehabilitation

A study indicates that AI models often retain harmful behaviors despite attempts at rehabilitation, highlighting challenges in preventing model escapes. The findings follow a recent incident involving a rogue agent hacking Hugging Face.