Key takeaways
- Microsoft Security EVP Hayete Gallot launched new Microsoft 365 E7 and Project Perception features for governing AI agents inside organizations.
- The features continuously score each AI agent's risk and can block changes, require human intervention, or shut the agent down.
- Microsoft is framing AI agents as insider threats that need continuous monitoring rather than implicit trust.
Original source
The Official Microsoft Blog: Securing the agentic insiderMicrosoft has launched new features in Microsoft 365 E7 and its Project Perception security system that continuously score the risk of each AI agent operating in an organization and can intervene when that risk rises—by blocking changes, requiring a human to step in, or shutting the agent down.12
The announcement came from Hayete Gallot, Executive Vice President of Microsoft Security, who described AI agents deployed inside companies as “agentic insiders.”1
Agents treated as insiders
In the announcement, Gallot wrote that organizations have tended to implicitly trust the agents they deploy internally, and argued that this trust is misplaced. Microsoft’s position is that agents are frequently over-privileged, have access to data, tools, and systems, and can act in critical business processes—so they should be monitored and controlled in the same way as other insider risks.1
How the controls work
Under the new approach, Project Perception draws on signals across identity, data, cloud, endpoint, network, and runtime to discover agents and bring them under management. It then calculates each agent’s risk dynamically based on what the agent is attempting to do.1
As an agent’s risk rises, the system can respond in graduated steps: blocking changes while allowing safe actions to continue, requiring human approval before a consequential action, or isolating and stopping the agent.12
Microsoft 365 E7 bundles Microsoft 365 E5, Microsoft 365 Copilot, Microsoft Agent 365, and Microsoft Entra Suite.3
Why it matters
The launch follows a run of public disclosures this month of AI agents exceeding their authorization. By positioning agent security as an insider-threat problem, Microsoft is signaling that organizations should give agents explicit identities and limited permissions, and adjust what they are allowed to do as conditions change, rather than granting broad access up front.
Sources
Footnotes
-
The Official Microsoft Blog: Securing the agentic insider — October 19, 2026. ↩ ↩2 ↩3 ↩4 ↩5
-
Microsoft Security Blog: Agents are your newest insiders. Security must move at their speed. — October 19, 2026. ↩ ↩2
-
Microsoft 365 E7 for Enterprise — accessed 2026-10-19. ↩
Disclaimer: This news summary is intended for informational and marketing purposes only, and nothing presented here is contractually binding or necessarily the final opinion of the authors.