Microsoft extends Project Perception to treat AI agents as insiders, with a permission set up per task and a risk score that changes

Microsoft said on 19 October that it is extending Project Perception and Microsoft Agent 365 to secure what it calls the agentic insider. Discovery grows by 35 types of local agents, to 60 in total, including Claude Code, OpenAI Codex and GitHub Copilot, and each agent gets its own identity through Microsoft Entra Agent ID. New Dynamic Permissions start an agent with no authority and grant only what an approved task needs, for as long as it needs it. Dynamic Agent Risk and Dynamic Conditional Access then raise or lower that access as the agent's behaviour changes.

• Write a one-page list of every AI tool at your business that can read email, files or accounts, and who in the team owns each one.
• For each agent, decide which actions it may take alone and which need a named person to approve, and switch on approval for anything that moves money or sends messages.

Source: Microsoft Security Blog. This is our short summary of their reporting. The paragraphs marked "Our read" and the things to do are SeedFoundry's opinion. The summary is drafted with AI tools. Image: Microsoft Security Blog, from its own announcement. How we source and credit