Artificial intelligence agents are becoming more autonomous and deeply embedded across enterprise environments, creating security risks that traditional access controls, endpoint tools and network defenses cannot fully address. Their non-deterministic behavior, opaque reasoning and ability to act beyond intended tasks make runtime oversight critical. This session covers how organizations can evaluate agent intent, access and behavior together, strengthen identity attribution and auditability, and use an agent security maturity model to assess existing capabilities and progress toward continuous runtime enforcement.

In this session, Jessica Marie of Proofpoint will discuss:

• Why semantic privilege escalation can occur even when every agent permission and authorization check is valid;
• Where traditional endpoint, network and firewall controls lose visibility across multi-step agent workflows;
• What foundational questions organizations should answer to demonstrate effective oversight to stakeholders and regulators.