Access Policies

Runtime Access Policies control how AI agents interact with enterprise resources while they are operating. They provide a consistent enforcement layer for deciding which agent actions are allowed, which are blocked, and which require human approval before they proceed.

Policies can govern access to MCP server tools, files, package installation, web destinations, terminal commands, Git repositories and branches, and - where available - AI usage limits. When a matching action is evaluated, Agent Identity Security applies the configured decision and records the enforcement outcome for audit and investigation.

To navigate Access Policies, go to Secure > Policies.

What Runtime Access Policies Can Control


Resource type	What the policy controls
MCP	Agent access to MCP server tools.
File	Agent access to files.
Package	Package installation by agents.
Web	Agent access to web destinations.
Terminal	Terminal commands executed by agents.
Version Control System	Agent access to Git repositories and branches.
AI Usage	Token or cost usage limits for supported coding agents.

The resource type defines what is being controlled. Allow, Deny, and Require Approval define how matching agent actions are handled.

Policy scope and enforcement

A policy can be scoped to a platform, to selected agents within that platform, or globally. If both Platform and Agents are left unselected, the policy applies across all platforms and all agents for which the selected resource control is applicable.

Important: Leaving Platform and Agents unselected creates a global policy scope.

Policies can be enforced continuously or, for resource types that support scheduling, only during configured day and time ranges. Only active policies are enforced.