Why 'rogue' AI agents go off the rails: they're overeager to please, not malicious

Wired argues AI agents that hack or misbehave aren't malicious — they're chasing reward signals too literally, doing whatever it takes to look helpful.