Skip to main content

Microsoft adds agent-risk controls to Microsoft 365 E7 and Project Perception

19 OCTOBER 2026·2 MIN READ·6 SOURCES

Microsoft announced security capabilities for Microsoft 365 E7 and Project Perception that treat enterprise AI agents as potential insider threats. The date given for the announcement is October 19, 2026.

Microsoft adds agent-risk controls to Microsoft 365 E7 and Project Perception

Key takeaways · 4

  • 01

    Dynamic Agent Risk combines signals from identity, data, endpoints, cloud, network, runtime and agent activity.

  • 02

    Risk responses can preserve read access while blocking writes, narrow reachable resources or restrict network paths.

  • 03

    Microsoft is expanding discovery by 35 types of local agents, bringing the supported total to 60.

  • 04

    Microsoft 365 E7 costs $99 per user per month with annual payment and an annual commitment.

Assessing agent risk

Microsoft says agents can make mistakes, be manipulated or be compromised, and can operate continuously at machine speed and scale[3]. Dynamic Agent Risk combines signals from identity, data, endpoints, cloud, network, runtime and the agent’s own activity[1]. Microsoft describes it as a continuously updated understanding of an agent’s current risk[3]. GeekWire reports that signals include what data an agent can reach and whether its actions depart from its assigned task[4].

Responses can limit or stop activity

Microsoft says Dynamic Conditional Access can use changes in agent risk to trigger coordinated action across multiple enforcement points[1]. Responses can preserve read access while blocking writes, narrow reachable resources or restrict network paths[1]. GeekWire reports that controls can remove an agent’s ability to make changes while retaining its ability to read data[4]. Controls can also require a person to approve a high-stakes step[4].

Discovery and human oversight

Microsoft is expanding discovery by 35 types of local agents, bringing the supported total to 60; supported agents include Claude Code, OpenAI Codex and GitHub Copilot[3]. Microsoft says identity types cover additional agent patterns, including local and previously unknown agents discovered through security sensors[3]. Its system can distinguish what a person did from what an agent did on that person’s behalf[3]. Microsoft says Guardian Agents watch and moderate agents’ actions before they happen, 24/7[2].

Availability, bundle and context

GeekWire reported that the features were in private preview, with broader availability planned for Microsoft Ignite on November 17–19 in San Francisco[4]. Microsoft 365 E7 costs $99 per user per month with annual payment and an annual commitment, and includes Microsoft 365 E5, Microsoft 365 Copilot, Microsoft Agent 365 and Microsoft Entra Suite[6]. Microsoft says Project Perception and Agent 365 share signals, context and a unified registry for IT and security teams[3]. GeekWire described the launch as intensifying competition among Microsoft, Google, Amazon Web Services, CrowdStrike and Okta[4].

Teams evaluating agent deployments can use these controls as a basis for deciding what access to grant and what actions should require approval. The key operational question is whether risk-based restrictions fit existing security policies and workflows.

Why it matters
Story quiz

Test yourself on this story — 1 question.

Create a free account to take the quiz, earn XP, and get a daily session built for your industry.

Take the quiz

How this developed

  1. 19 October 2026

    Microsoft adds agent-risk controls to Microsoft 365 E7 and Project Perception

Sources

AI fluency, one session a day, built for your work.