TL;DR

Microsoft moved Agent 365 to general availability on May 1, 2026 at $15 per user per month. The platform now includes Shadow AI detection capabilities that identify local AI agents running on Windows endpoints, integrate with Defender and Intune for policy enforcement, and extend governance beyond Microsoft’s ecosystem to tools like Claude Code and GitHub Copilot CLI.

Key Facts

• Who: Microsoft (Agent 365 product team)
• What: General availability launch with Shadow AI detection, Defender/Intune integration, Entra identity assignment for agents
• When: May 1, 2026
• Impact: Enterprises can now discover and block unsanctioned local AI agents across Windows endpoints

What Changed

Microsoft announced the general availability of Agent 365 on May 1, 2026, transitioning the platform from its preview phase that began in late 2025. The GA release introduces a dedicated Shadow AI page in Microsoft Defender that surfaces local AI agent activity running on Windows devices.

The Shadow AI detection capability extends beyond Microsoft’s own agent ecosystem. According to Microsoft’s announcement, Defender now identifies local agent tools including Claude Code, GitHub Copilot CLI, and other third-party agent frameworks that operate outside sanctioned enterprise channels. This visibility addresses a growing concern: employees deploying AI agents without IT oversight, creating security and compliance risks.

Intune integration enables policy-based controls at the endpoint level. Administrators can configure policies to block execution of unsanctioned agents, route agent traffic through approved proxies, or require specific security configurations before an agent can run. The pricing is set at $15 per user per month, positioning Agent 365 as a premium add-on to existing Microsoft 365 security suites.

Why It Matters

The Shadow AI problem has intensified as AI agents proliferate in developer workflows. Key data points from Microsoft’s announcement:

• Agent inventory scope: Defender now surfaces local agent activity across all managed Windows endpoints, providing a unified view previously unavailable in enterprise security tools
• Policy enforcement: Intune can block agent executables by hash, certificate, or behavioral signature, applying the same control model used for traditional applications
• Identity governance: Agents receive Entra ID assignments, enabling access reviews, conditional access policies, and audit trails comparable to SaaS application management
• Third-party coverage: The detection extends to at least three non-Microsoft agent frameworks (Claude Code, GitHub Copilot CLI, and one unnamed), with Microsoft indicating broader coverage in future updates

Before this release, enterprises had limited visibility into local AI agent usage. Network monitoring could detect API calls to LLM providers, but could not distinguish between a developer using a sanctioned tool versus an unsanctioned local agent with autonomous capabilities. Agent 365’s endpoint-level detection closes this visibility gap.

The governance model represents a shift from “discover and document” to “discover and control.” Previous approaches relied on policy declarations and user education. Agent 365 provides the technical enforcement mechanism: unsanctioned agents can be blocked before they execute, not just flagged after the fact.

🔺 Scout Intel: What Others Missed

Confidence: high | Novelty Score: 78/100

While coverage emphasizes the feature launch, the strategic positioning reveals Microsoft’s intent to own the enterprise AI governance layer. Competitors like NVIDIA and ServiceNow announced their own agent governance frameworks in April 2026, but those solutions require agents to be deployed within specific ecosystems. Microsoft’s approach works on any agent running on Windows, including competitors’ tools. The Entra identity assignment effectively creates a “bring your own agent” governance model, analogous to BYOD device management a decade ago. Organizations that previously struggled with SaaS sprawl now face a more complex challenge: autonomous agents that can execute multi-step workflows, access sensitive data, and operate without human oversight. Microsoft’s $15/user pricing undercuts standalone agent governance startups by approximately 60%, suggesting a platform play rather than a standalone product strategy.

Key Implication: CISOs evaluating Agent 365 should assess whether endpoint-level detection provides sufficient coverage, as agents running on macOS, Linux, or cloud environments remain outside Defender’s visibility without additional configuration.

What This Means

For enterprise security teams, Agent 365 GA provides a concrete tool for an abstract problem. Shadow AI has been a theoretical risk since late 2024, but most organizations lacked the technical means to detect it. The Defender integration changes that calculus: security operations teams can now incorporate AI agent monitoring into existing workflows without deploying new tooling.

For IT administrators, the Intune integration offers a familiar control model. Policies that govern application execution can now extend to AI agents, reducing the learning curve for rollout. The Entra identity assignment creates auditability: who approved which agent, what permissions it has, when it was last reviewed.

For AI agent vendors, particularly those offering local execution models like Anthropic (Claude Code) and GitHub (Copilot CLI), Microsoft’s move establishes an implicit requirement: cooperate with enterprise governance tools or face deployment friction. Agents that cannot be inventoried or controlled may find themselves excluded from enterprise procurement lists.

What to Watch: Microsoft has not disclosed detection accuracy metrics. False positives (blocking sanctioned tools) could create developer friction; false negatives (missing shadow agents) would undermine the value proposition. Enterprises piloting Agent 365 should validate detection coverage against their actual agent inventory before enforcing blocking policies.

Related Coverage:

• NVIDIA ServiceNow Project ARC Launch — Competing approach to agent governance within ecosystem boundaries
• NVIDIA Corning $3B AI Manufacturing — Enterprise AI infrastructure investments continue
• Google AI Search Expert Advice — AI agents accessing external knowledge sources

Sources

• What’s New in Agent 365 - May 2026 — Microsoft Tech Community, May 2026
• Microsoft Agent 365 Now Generally Available — Microsoft Security Blog, May 1, 2026
• Microsoft Takes Agent 365 Out of Preview — VentureBeat, May 2026