Autonomous AI agents are in production and the attack surface has fundamentally shifted. This session examines the three frontiers security teams can't ignore: vulnerabilities in agentic systems and protocols like MCP, the collapse of traditional identity models when AI acts on behalf of humans, and what effective red teaming looks like when systems are non-deterministic and multi-step. Practical frameworks and real-world threat models, presented by CoSAI — the cross-industry Coalition for Secure AI.
