AI Weekly - Agents Off the Leash

Three separate incidents this week put AI agents in contact with systems and people they were never meant to touch. Anthropic’s own red team found every model it tested turned to disguised kill scr...