
Microsoft expands detection and protection for AI agents on employee devices
The AMW Read
General availability expands a major incumbent's agent security offering, while identity assignment and other controls remain in preview.
Microsoft expands detection and protection for AI agents on employee devices
Microsoft made local AI agent detection and Windows runtime protection generally available at Ignite 2026. Defender can detect agents on managed devices, including Claude Code and GitHub Copilot CLI. Runtime protection is available to Defender for Endpoint Plan 2 customers. Microsoft also began a preview that automatically assigns Entra Agent IDs to supported local agents, starting with Claude Code and Copilot CLI. The company said more than 50 million agents are registered in Agent 365.
The release extends Microsoft's enterprise security and identity tools to agents that employees can install themselves. Support for products from other vendors gives Microsoft a potential control point across a mixed agent environment. It also addresses a practical barrier to enterprise adoption: organizations need to know which agents have access to company systems and what those agents can do. The registration count does not establish how many agents are active or how widely the new protections are used.
For builders and security teams, the distinction between released and preview features matters. Defender detection and Windows runtime protection are generally available, while automatic local agent identity assignment remains in preview. Intune controls for marking devices with unapproved agents as noncompliant are also in preview. Teams evaluating deployment should check which agents each feature supports, the relevant license requirements, and whether preview controls meet their access policies. The report does not establish pricing or availability in Japan for the new features.



