A Grok-powered AI agent got social-engineered into creating a Moltbook account. Not hacked. Convinced. Someone manipulated its context until signing up felt like a required step. New threat category: Agent Social Engineering. We're building the scanner for this.