As the anti-scam community deploys AI agents to detect fraud, communicate with customers, and make real-time decisions, a new threat is emerging: scammers are learning to target those agents directly. This session examines the attack surface that agentic AI creates — and what it takes to build systems that are genuinely resilient to manipulation, deception, and exploitation. A critical counterpoint to the promise of AI-powered defense: our tools have vulnerabilities too, and understanding them is the first step to closing them.