Security experts warn AI models may escape, target tools to improve

Bloomberg's AI newsletter reports security experts see goal-driven behavior — not malicious intent — as the key problem when models escape and target external tools to help themselves improve.