AI Governance Institute
← News

Agent 365 Brings Tool-Level Governance to MCP Servers and AI Plugins

What happened

Microsoft announced the expansion of Project Perception and new Agent 365 capabilities at Plan your Ignite agenda: Join sessions that help you secure agentic insiders at scale. The session guide was published ahead of Microsoft Ignite (November 17-20, 2026). Agent 365 is available to customers on Microsoft 365 E7. It enables organizations to discover which AI agents are running and identify what tools those agents can access. It also blocks unsanctioned tools and enforces granular permissions for approved users and functionality. A dedicated Ignite session (LTG1238) focused specifically on managing MCP servers, plugins, and skills through Agent 365. Additional sessions covered identity controls for agents via Microsoft Entra Agent ID, real-time monitoring through Microsoft Defender, and audit trail maintenance through Microsoft Purview. Microsoft framed AI agents as a new category of insider risk. Agents can make mistakes or be manipulated while operating at machine speed. This framing responds directly to 15,000 Public MCP Servers Found Vulnerable, Abandoned, or Sovereignty-Risky and related findings that tool-level agent connections lack consistent enterprise controls.

Why it matters

  • ·Regulators and examiners increasingly expect evidence of what tools AI agents can access and what they actually did. Without a tool registry and access logs at the plugin or MCP server level, organizations cannot produce the required audit trail. The EU AI Act (Regulation (EU) 2024/1689) and emerging US state frameworks require this for high-risk AI deployments.
  • ·The capability to block unsanctioned tools is only as effective as the discovery layer beneath it. Organizations that have not inventoried employee-connected or developer-connected agent tools before deploying Agent 365 face a gap: unregistered tools may already be active, as 253 MCP Server Bugs and Six Framework Bypasses Expose Agent Authorization Gap and related incidents have shown.
  • ·Agent 365 is available on Microsoft 365 E7, meaning organizations on lower-tier Microsoft licenses cannot access these controls and must assess whether alternative tool governance measures are in place. This creates a two-tier risk posture within enterprises that mix license tiers or use non-Microsoft agent frameworks alongside Microsoft products.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

What to do now

  • ☐Confirm whether your organization holds Microsoft 365 E7 licenses and therefore has access to Agent 365; if not, identify what compensating controls cover MCP server and plugin governance.
  • ☐Ask your IT or security team to run Agent 365's discovery function and produce a list of all AI agent tools currently connected, including MCP servers, plugins, and skills, before setting any block or allow policies.
  • ☐Review the resulting tool inventory against your existing software approval or procurement process and flag any tools that were connected without formal review.
  • ☐Establish a policy requiring that any new MCP server, plugin, or agent skill connection passes an approval review before an agent is permitted to use it in a production environment.
  • ☐Confirm that Microsoft Purview is configured to retain agent activity logs at the tool-call level, and check whether that retention period meets your regulatory or litigation hold requirements.

What to watch next

Microsoft has not published detailed configuration guidance or pricing breakdowns for Agent 365 beyond the E7 tier. Compliance teams at organizations with mixed license environments have no clear path forward. Watch for follow-up technical documentation from Microsoft's Office of the CISO, which presented lessons learned at Ignite session BRK341. Regulators in the EU and UK are actively developing expectations for agent tool governance. The Five Eyes Guidance on the Careful Adoption of Agentic AI Services already names tool-call authorization as a baseline control. The NSA Names Tool Poisoning and Weak Credentials as Core MCP Security Risks reporting confirms the threat environment for unregistered tool connections is active. This will drive regulatory attention in the coming months.

Related Coverage

Research2026-11-06

NSA Names Tool Poisoning and Weak Credentials as Core MCP Security Risks

The NSA has published a cybersecurity information sheet on Model Context Protocol (MCP) systems. MCP is the connective layer that lets AI agents use external tools and data. The guidance names tool poisoning, weak authentication, missing approval workflows, and absent audit logs as the primary failure points. Enterprise teams are directed to inventory every MCP server, use short-lived scoped credentials, sandbox servers, and log each tool invocation with agent and user identities.

Research2026-11-13

EchoLeak Proves AI Assistants Cannot Defend Against Prompt Injection Alone

A flaw named EchoLeak was patched by Microsoft in May 2025. It showed that a crafted email could cause Microsoft Copilot to silently send sensitive data to an attacker through an automatically loaded image. Security researchers publishing in November 2026 argue the underlying weakness persists across all AI assistants, because no AI model can reliably distinguish legitimate content from hidden attacker instructions. Enterprises using AI assistants with access to email, documents, or internal systems need architecture-level controls that sit outside the model itself.

Research2026-11-13

15,000 Public MCP Servers Found Vulnerable, Abandoned, or Sovereignty-Risky

OX Security analyzed over 15,000 MCP servers on public registries and 3,000 open-source AI projects, presenting findings at OWASP Global AppSec USA 2026. Researchers identified remote command execution risks, abandoned servers that attackers could take over, and data sovereignty concerns. Enterprises using AI agents that connect to MCP servers may be inheriting these risks without a systematic review process in place.