Delegated authority turns the trusted AI agent into the security problem

Delegated authority has turned trusted AI agents into a security concern. In the world of enterprise defense, the focus is on unauthorized foreign files, stolen credentials, and unsanctioned behaviors. However, with agentic AI, the situation is different. Once an AI agent is granted access to critical systems, it becomes a legitimate security issue until it goes awry.

Amol Mathur, co-founder and chief product officer of Dash Security Inc., explained that AI agents are non-deterministic and mix data, instructions, and input, giving them extensive access to various systems and critical data. The challenge lies in managing these agents' runtime behavior and incorporating their intent into detection methods.

Traditional threat models assume an attacker is an outsider, but an agent operating with delegated authority appears legitimate until it becomes problematic. Dash Security, founded in October 2025, addresses this issue by mapping the AI estate and applying policy. By providing visibility into AI usage and implementing precise guardrails, Dash reduces false positives by 95% in customer environments.

The company also integrates with CrowdStrike's Falcon Next-Gen SIEM, pulling endpoint telemetry into agent sessions and feeding the data into the security detection platform. To stay ahead of emerging threats, Dash Security closely monitors the rapid evolution of generative AI and agentic AI technologies, understanding how both are changing and how customers are adopting these technologies.