Release Notes for Agent Identity Security

Discover coding agents running on enterprise endpoints, including GitHub Copilot, Claude Code, and Cursor. Discover enterprise AI agents through integrations with Microsoft Copilot Studio, Amazon Bedrock AgentCore, Amazon Bedrock, and n8n, and bring them into the Agent Identity Security inventory.

Detect Shadow AI agents through integration with CrowdStrike Falcon or through the Agent Identity Security discovery agent deployed to Windows endpoints using Microsoft Intune. Gain visibility into short-lived scripts, browser-based AI agent activity, and interactions with files and MCP servers, with the ability to disable Shadow AI agents using Kill Switch controls.

Discover and register custom and enterprise-developed AI agents using the Agent Identity Security Developer SDK. Bring these agents into Agent Identity Security for centralized visibility and governance.

Integrate with enterprise DLP and data classification platforms to bring file classification context into Agent Identity Security. Use classification information in Runtime Access Policies to control agent access to sensitive files.

Integrate with Microsoft Entra ID to discover users and groups and bring enterprise identity context into Agent Identity Security. Associate users and groups with AI agents to provide greater visibility into agent identities, access, and relationships.

Integrate with enterprise communication and collaboration platforms to support Human in the Loop approval workflows. Route agent actions that require approval to designated approvers, allowing the action to proceed only after approval.

Maintain a centralized AI Bill of Materials (AIBOM) for discovered agents across platforms. View agent ownership, status, risk, activity, associated MCP servers, models, skills, users, and credentials, visualize access relationships through the Access Graph, and review agent activity over time. For Claude Code agents, view token consumption and usage information directly from the agent's Usage tab.

Discover AI skills associated with agents and gain visibility into their usage across the organization. Sanction or unsanction skills based on organizational requirements.

Activate or deactivate AI agents directly from the Agent Registry using Kill Switch controls. Track Kill Switch actions through Event Logs.

Route agent communication with MCP servers through the MCP Gateway and apply granular access controls at the server and tool level. Support OAuth, API keys, bearer tokens, and PAT pass-through, with Just-in-Time (JIT) access enabled through OAuth, Human in the Loop approvals, and controls to prevent direct access to governed MCP servers.

Inspect requests and responses between AI agents and MCP servers for PII, secrets, sensitive data, and prompt injection. Apply policies to allow, block, or redact sensitive data as it passes through the MCP Gateway.

Define and enforce Runtime Access Policies to control what AI agents can access and the actions they can perform during execution. Apply granular controls across terminal commands, websites, version control, files, packages, and MCP resources, with the ability to allow, warn, block, or require Human in the Loop approval. Define token usage limits for Claude Code agents to control token consumption at runtime. Runtime Access Policies can also use data classification information from enterprise DLP integrations to restrict agent access to sensitive files.

Define Intent Based Access Policies to evaluate whether an agent's actions remain aligned with the user's original intent as a task is executed. When intent drift is detected, the action can be logged and allowed to continue or blocked from execution, with policy decisions captured in Event Logs.

Assess AI agents against security posture checks to identify security and governance risks. View posture issues, understand why they were identified, and track them as issues are resolved.

Track agent compliance across 10 compliance frameworks, including SOC 2, ISO 27001, NIST, GDPR, and HIPAA. Identify failing controls and the agents associated with compliance violations.

Gain a centralized audit trail of agent activity, MCP Gateway communication, runtime policy decisions, security events, and governance actions. Search and filter events for investigation and audit.

Use out-of-the-box workflows to automate agent security and governance processes. Trigger workflows based on agent and security events, add Human in the Loop approvals, and automate actions such as enabling or disabling agents, sending notifications, and creating tickets.

Integrate Agent Identity Security with enterprise identity providers for authentication and user provisioning. Support multiple identity providers and synchronize users and groups through SCIM.

Control access to Agent Identity Security capabilities based on user roles and responsibilities. Use predefined roles including System Admin, System Admin Read Only, Security Engineer, and GRC Analyst to provide appropriate levels of access.

Access Agent Identity Security capabilities through an MCP client without using the product interface. Query Agent Identity Security data through MCP while applying platform authentication and access controls.

Use natural language to investigate Agent Identity Security and get insights across agents, risks, events, policies, compliance, and MCP Gateway. Navigate directly to relevant product areas and get guidance on Agent Identity Security capabilities from the assistant.