AI Agent Security: The Lethal Trifecta of Vulnerabilities in Enterprise Deployment

Ian Webster, CEO of Promptfoo, discusses the emerging security challenges of AI agents in enterprise environments. He outlines the "Lethal Trifecta" framework identifying three critical vulnerability conditions: untrusted user input, access to sensitive information/PII, and outbound communication channels. The conversation emphasizes that agent security requires integration across identity, API security, and AI layers, and that security testing should shift left into developer workflows rather than remaining a late-stage gate.