StrongDM has developed an open-source project called Leash, designed to improve security for AI agents operating in enterprise environments. As organizations increasingly deploy AI agents that execute tasks autonomously, the need for robust access controls becomes critical. Leash extends StrongDM's existing access control framework to these non-human actors, allowing teams to monitor and govern their behavior in real-time. It employs a policy language called Cedar, enabling organizations to define clear rules that agents must follow when interacting with systems and data. Leash operates at a low level within the network stack, inspecting outbound traffic and enforcing policies before connections are established. It provides context-aware rules based on various factors, like the agent's identity and the destination attributes. This approach prevents unauthorized actions and ensures all agent activities are logged for accountability. The system is designed to be lightweight, adding minimal performance overhead while scaling across container environments. In addition to monitoring agent behavior, Leash also integrates with the Model Context Protocol (MCP) to control what information AI agents can access. By parsing MCP calls, it helps block unauthorized access and creates auditable logs, allowing organizations to respond quickly to potential threats. Overall, Leash represents a significant shift in how access control is managed, transitioning from traditional human-centric models to a more comprehensive framework that includes both human and AI agents.