Security

Microsoft Agent 365 GA Adds Security and AI Controls

3 min read

Summary

Microsoft Agent 365 is now generally available for commercial customers, giving IT and security teams a unified control plane to observe, govern, and secure AI agents across Microsoft 365, endpoints, and cloud environments. New preview capabilities also extend visibility to shadow AI, local Windows agents, multicloud agent platforms, and policy-based controls through Defender and Intune.

Need help with Security?Talk to an Expert

Microsoft Agent 365 reaches general availability

Introduction

AI agents are quickly moving from pilot projects into daily business operations, often with access to sensitive data, tools, and workflows. Microsoft Agent 365 is now generally available for commercial customers, giving organizations a central way to monitor, govern, and secure these agents before agent sprawl turns into a security problem.

What’s new in Microsoft Agent 365

Microsoft positions Agent 365 as a control plane for AI agents across Microsoft and partner ecosystems. With general availability, organizations can now manage:

  • Agents acting on behalf of users with delegated access
  • Agents operating independently with their own credentials and permissions
  • Agents participating in team workflows in public preview

The release also adds several important preview capabilities:

Shadow AI and local agent discovery

Through Microsoft Defender and Intune, organizations can discover and manage local AI agents on Windows devices. Initial support includes OpenClaw, with expansion planned for tools such as GitHub Copilot CLI and Claude Code.

Admins can:

  • Identify where local agents are running
  • See discovered agents in the new Shadow AI experience
  • Use Intune policies to block common ways unmanaged agents run

More context for security investigations

Starting in June 2026 public preview, Defender will add richer asset context mapping for agents, including:

  • The devices where agents run
  • Configured MCP servers
  • Associated identities
  • Reachable cloud resources

This should help security teams understand exposure, investigate suspicious activity, and prioritize high-risk agents.

Runtime protections and policy controls

Microsoft also announced policy-based controls and runtime protections for coding agents. If an agent shows malicious behavior, such as attempts to access or exfiltrate sensitive data, Defender will be able to block the agent and raise alerts with investigation context.

Multicloud visibility

Agent 365 now supports public preview registry sync with AWS Bedrock and Google Cloud, helping admins discover and inventory cloud agents across multiple AI builder platforms. Microsoft also plans to support basic lifecycle actions such as starting, stopping, and deleting agents.

Why this matters for IT and security teams

For administrators, the main value is visibility and consistency. AI agents increasingly exist outside traditional governance processes, whether on endpoints, in SaaS apps, or across cloud platforms. Agent 365 brings those signals into familiar Microsoft 365, Intune, and Defender workflows.

This is especially important for:

  • Controlling shadow AI usage
  • Reducing overprivileged agent access
  • Investigating agent-related incidents faster
  • Applying governance across hybrid and multicloud environments

Next steps

IT and security teams should review current AI agent usage across Microsoft 365, Windows devices, and connected cloud platforms. Organizations already using Intune and Defender should also evaluate the new preview capabilities, especially for local agent discovery, policy enforcement, and multicloud registry sync.

As AI adoption accelerates, Agent 365 gives Microsoft customers a more practical way to secure agents without introducing a separate management model.

Need help with Security?

Our experts can help you implement and optimize your Microsoft solutions.

Talk to an Expert

Stay updated on Microsoft technologies

Read the original article by Nirav Shah, Rob Lefferts and Jason Roszak
Microsoft Agent 365AI agentsMicrosoft DefenderIntuneshadow AI

Related Posts

Security

Microsoft Defender Email Security Benchmark Insights

Microsoft has shared one year of real-world email security benchmarking data showing Defender consistently leading in pre-delivery detection versus SEG vendors. The latest results also show ICES tools add the most value for promotional and bulk email, while Defender now handles the vast majority of post-delivery malicious remediation.

Security

ASSERT Framework Turns AI Specs Into Executable Evals

Microsoft has released ASSERT, an open-source framework that converts natural-language behavior requirements into executable evaluation pipelines for AI models, agents, and applications. The tool helps teams build behavior-specific tests faster, improve regression coverage, and better validate whether AI systems follow product policies and safety expectations.

Security

AI Activity Investigations: New Microsoft Playbook

Microsoft has published a new investigator playbook to help security teams reconstruct AI-related activity across Microsoft 365 Copilot and Azure AI services. The guidance brings together telemetry, KQL queries, schema references, and detection logic across Purview, Defender, and Sentinel so investigators can move from isolated signals to a clear incident timeline.

Security

AI Brand Phishing Campaigns Target Microsoft Users

Microsoft Threat Intelligence reports a rise in phishing, malvertising, and SEO-driven attacks that abuse popular AI brands like ChatGPT, Claude, Copilot, and DeepSeek as social engineering lures. The campaigns use familiar tactics such as urgent payment notices, fake policy violations, and malicious installers to steal credentials, payment data, and deploy malware, making user awareness and layered defenses critical.

Security

AI GitHub Actions Secret Exposure in Claude Code

Microsoft Threat Intelligence found that Anthropic’s Claude Code GitHub Action could expose CI/CD secrets when AI agents process untrusted GitHub content such as issues, pull requests, and comments. Anthropic fixed the issue in Claude Code 2.1.128, but the research highlights broader risks for any AI-enabled workflow with access to secrets, file reads, or outbound communication.

Security

Agentic AI Failure Modes Taxonomy Updated by Microsoft

Microsoft has updated its taxonomy of failure modes in agentic AI systems after a year of red teaming against real-world deployments. The v2.0 framework adds seven new risk categories and expanded mitigations, giving security teams a more practical model for assessing agentic AI threats such as MCP/plugin abuse, goal hijacking, and session context contamination.