AI chatbots tested for phishing risks: volunteers received scam-like messages, exposing gaps in safety controls.