Microsoft adds security checks for Copilot agents
Microsoft Defender now evaluates security posture for agents built in Copilot Studio and Foundry. It flags excessive permissions, misconfigurations, and potential attack paths. The tool supplies prioritized recommendations and attack path analysis. It requires the Global Secure Access client for device-level agents and focuses on prompt injection risks plus URL filtering. The update targets companies building no-code agents in SharePoint and Teams.
Before this change, teams could spin up agents with little visibility into permission sprawl or data exposure risks. Security reviews happened after problems surfaced or not at all. Now security teams gain structured risk context before agents reach users. That raises the bar for approval and makes casual experimentation more visible to IT, which slows the very pilots mid-size teams need to prove value.
Analysis
Stop treating agent building as low-risk tinkering. Run one narrow pilot with explicit security review upfront, document the permission model, and show the risk controls to your boss before asking for broader access.
Pulse published by Collab365 Spaces. Cite as "Microsoft adds security checks for Copilot agents", Collab365 Spaces. 1 source referenced.