Does Sentinel cover AI governance and AI observability?

Yes - that's the AI governance suite in the Enterprise edition. For governance, it inventories every Copilot Studio agent (including shadow agents nobody registered), reviews each agent's Entra ID permissions, scores agent risk on six factors, and lets admins block users, quarantine an agent, or disable its identity. For observability, it tracks agent conversations, outcomes, and errors, classifies every user prompt for jailbreak and injection attempts, and measures Microsoft 365 Copilot adoption.

Where does Sentinel run, and where does our data live?

In your own tenant. Sentinel deploys into your Azure subscription - the web app, the worker, and a PostgreSQL database that holds the synced Microsoft 365 inventory, with configuration in your Key Vault. Sign-in runs through your Microsoft Entra ID. There is no shared multi-tenant backend, and your data is never used to train public models.

What permissions does it need to connect?

An Entra app registration with read-only Microsoft Graph permissions to read your directory, SharePoint, OneDrive, sharing, and Copilot usage reports, plus read access to the Application Insights resource your Copilot Studio agents report to. Remediation - revoking links, blocking users, quarantining agents - uses separately scoped permissions and only runs when an admin starts it.

Does Sentinel change anything in our environment on its own?

No. Syncing, scoring, and prompt classification are read-only. Every state-changing action - revoking a link, a bulk revoke, blocking a user from an agent, quarantining an agent, or disabling its identity - is admin-initiated and written to an audit log with actor, target, outcome, and timestamp.

How are risk scores calculated?

Deterministically, from real exposure signals. Data risk scores every site and OneDrive on five factors: exposure from anonymous and organization-wide links, external reach, blast radius, dormancy, and link hygiene. Agent risk scores every custom Copilot Studio agent on six: reach, data access and tooling, safety threat exposure, operational failures, usage, and control and hygiene. The breakdown is always shown, and an optional AI layer writes a plain-language summary on top.

Can we tune what counts as an alert?

Yes. Sentinel ships with 34 built-in alerts across data, agents, AI Safety, and Copilot. Admins set the Warning and Critical thresholds for the 13 data alerts - for example how many anonymous links is too many, or how long before a guest counts as inactive - and every change is audited.

Can it send reports to people who never log in?

Yes. An Executive Summary PDF and a Risk Summary PDF can each be emailed to their own recipient list on a daily, weekly, or monthly schedule, from a mailbox in your tenant. Recipients are limited to domains you approve, and personal mail providers are refused.

What does Sentinel need - do we have to be on Microsoft 365 E5?

Sentinel works against standard Microsoft 365 SharePoint, OneDrive, and Entra ID. Agent telemetry and AI Safety need your Copilot Studio agents connected to Application Insights, and Copilot Utilization depends on the Microsoft 365 Copilot usage reports your licensing exposes. Bring your environment to the demo and we'll confirm exactly what's covered.

Can we delegate cleanup to department owners instead of doing it all centrally?

Yes - that's what the Data Steward role is for. Members of a data-steward Entra ID group get a scoped 'Data Stewards Dashboard' that shows only their own department's oldest shared files, with the sharing evidence for each, so they can remediate their team's exposure without ever touching the full admin console. Central security stays focused on tenant-wide risk while data owners handle their own.

What roles and access levels are there?

Three Entra ID groups you control: administrators (full console, jobs, reports, and remediation), users (the posture, risk, and analytics dashboards), and data stewards (the department-scoped cleanup view). Access is least-privilege by default, membership is re-checked against Entra while people work, and you can optionally restrict sign-in to the countries you allow.

How is it licensed and deployed?

Per tenant, with a signed license file. Essentials covers sync, the dashboards, Data Risk, and email reports; Professional adds permission remediation; Enterprise adds the AI governance suite - agents, Agent Risk, AI Safety, quarantine, and Copilot Utilization. Trials include everything. It ships as containers on Azure Container Apps, so upgrades are a redeploy. Pricing is scoped to your environment - talk to us and we'll map it to your tenant.

See Sentinel on your own tenant

A 30-minute walkthrough on your data: exposure scores, agent risk, and a rollout plan for your Azure environment.

Request a demo