5.3.2. Plugins, Microsoft Agents, and Security Store Agents

💡 First Principle: Plugins are how Security Copilot reaches a data source, and agents are how it carries out a task on its own. The two are managed differently, and neither widens anyone's access to data.

Plugins connect Copilot to specific sources — Defender XDR, Sentinel, Entra, Intune, Purview, Defender Threat Intelligence, and others. Preinstalled Microsoft and non-Microsoft plugins are available to every owner and contributor by default; an owner can restrict any of them to owners only, which takes effect immediately in both the standalone portal and embedded experiences. Some preinstalled plugins, such as Microsoft Sentinel and Azure AI Search, also need setup (for Sentinel, which workspace to query) before they return data, configured per user or at workspace level. Custom plugins are owner-only by default until an owner lets contributors add them. Microsoft agents are pre-built, task-specific agents from Microsoft, such as the Phishing Triage Agent in Microsoft Defender. Security Store agents are third-party and partner-built agents acquired through the Microsoft Security Store. Unlike plugins, each agent is set up individually: it runs under its own identity, which you grant only the permissions its task needs, and the plugins it depends on are enabled for that agent alone.

⚠️ Exam Trap: An agent doesn't borrow the access of the person who enabled it. The Phishing Triage Agent runs as a dedicated agent identity and can classify alerts or read the related email content only if that identity has been granted those permissions — so "the agent can't read the emails" is an agent-identity permission question, not a plugin question.

Reflection Question: Why does Microsoft run an autonomous agent like the Phishing Triage Agent under its own identity instead of under the account of the administrator who turned it on?