Give an agent read access to a transaction database and tool access to a payments API, and you have created an attack surface nobody signed off on in the threat model. This session brings together practitioners securing agentic AI in live financial environments, covering how they scope permissions per tool call, manage credentials that agents use without a human in the loop, and contain blast radius when an agent takes an action it should not have been able to take. What this session will cover: - How to scope tool and data access at the agent level, not just the application level - Credential and secrets management for agents acting without human sign-off - Containing blast radius when an agent oversteps its intended scope - Balancing access controls against latency and throughput in live financial pipelines If you are responsible for the security posture of an agentic system touching financial data, this session is where you compare notes with people solving the same problem.