The definition

What is AI governance in Microsoft 365?

AI governance is the set of controls that keep AI accountable: knowing which AI agents exist, who uses them, what data and permissions they hold, how they behave, and how quickly you can intervene when one goes wrong. In Microsoft 365 that means governing two things at once - the Copilot Studio agents your teams build, and the SharePoint and OneDrive data that Copilot and those agents can reach.

01

Discover

Inventory every Copilot Studio agent - including shadow agents that are active but were never registered.

02

Assess

Score each agent for reach, data access, safety exposure, and hygiene, and review the Entra ID permissions it holds.

03

Control

Block a user, quarantine an agent, or disable its identity from one console - every action admin-initiated.

04

Prove

Audit logs, CSV exports, and scheduled executive reports show what was found and what was done.

Agent inventory

Find every agent - including the shadow ones

You can't govern what you can't see. Sentinel builds a custom agent inventory from live Copilot Studio telemetry and the agents registered in Dataverse, then flags shadow agents: active in conversations, but never registered.

  • Custom agent inventory - every Copilot Studio agent with its conversations, outcomes, response times, and tool success.
  • Shadow agent detection - agents receiving traffic that were never registered are surfaced for review.
  • Containment checks - agents quarantined but still receiving traffic, or disabled in Entra ID but still in use, are flagged.
  • Per-agent drill-down - topics, tools, errors, and daily risk history for any single agent.
app.princetonsentinel.com/dashboard/agents
Custom agent inventory listing Copilot Studio agents with conversation, outcome, and response-time metrics
Permissions & risk

Know what each agent is allowed to do

An AI agent is an identity with permissions. Sentinel reads each agent's Entra ID app registration - application permissions, delegated scopes, tenant-wide admin consent, and credential expiry - and folds it into an agent risk score alongside reach, safety exposure, and operational health.

  • High-privilege scopes - Graph permissions that can grant further access weigh more than broad read access.
  • Tenant-wide consent - admin-consented permissions that apply across the whole tenant are called out.
  • Credential hygiene - expiring app credentials are flagged before they turn into an outage or an incident.
  • Six risk factors - reach, operational failures, data access and tooling, safety threat exposure, usage intensity, and control and hygiene findings.
app.princetonsentinel.com/dashboard/agent-risk/sales-proposal-assistant
Agent risk detail for one Copilot Studio agent with exposure gauge, six-factor breakdown, and daily score history
Control

Contain a risky agent without filing a ticket

When an agent crosses a line, governance has to become action. Admins can block a specific user from a specific agent, quarantine the agent in Copilot Studio, or switch off its Entra identity - and reverse each once the review is done.

  • Per-user block and unblock - cut one person off from one agent, enforced through Dataverse.
  • Agent quarantine - take a Copilot Studio agent out of service through the Power Platform pending review, then release it.
  • Disable the agent's identity - turn off the agent's Entra service principal to stop its data access outright.
  • Every action logged - Agent Access and Agent Quarantine logs record who did what and when, with CSV export.
app.princetonsentinel.com/admin/agent-quarantine
Agent Quarantine log listing quarantine and release actions with admin, agent, reason, and status
Data governance for AI

Govern the data Copilot can reach

Copilot and Copilot Studio agents answer with whatever the user can already open. Every 'Anyone' link, organization-wide share, and forgotten guest is data your AI can surface. Sentinel scores every SharePoint and OneDrive site for that exposure and turns it into a revoke queue.

  • Oversharing found - anonymous links with no expiry, organization-wide links, external recipients, and guests.
  • Per-site data risk - a 0-100 score from exposure, external reach, blast radius, dormancy, and hygiene.
  • Bulk revoke - remove sharing links or non-owner access on selected files - admin-initiated and logged.
  • Department visibility - data stewards see their own department's oldest shared files, scoped by Entra ID.
app.princetonsentinel.com/dashboard/data-risk
Data risk page scoring SharePoint sites for oversharing exposure
Evidence

Show auditors and leadership what changed

Governance has to be provable. Sentinel keeps a record of every revoke, block, and quarantine, and puts the posture story in front of the people who need it, on a schedule.

  • Audit logs - revoke activity, agent access, and agent quarantine logs, each exportable to CSV.
  • Executive Summary - data, agent, and Copilot KPIs with active alerts in one view.
  • 34 built-in alerts - across data, agents, AI Safety, and Copilot, with admin-set thresholds for the 13 data alerts.
  • Scheduled PDF reports - Executive Summary and Risk Summary PDFs emailed daily, weekly, or monthly to the people you choose.
app.princetonsentinel.com/dashboard
Executive Summary with KPI cards, status bars, and urgent active alerts across data, agents, and Copilot
FAQ

AI governance questions

What is AI governance for Microsoft 365?

AI governance for Microsoft 365 is how an organization keeps Microsoft 365 Copilot and Copilot Studio agents accountable: an inventory of every agent, a view of the permissions and data each one can reach, ongoing risk and safety monitoring, the ability to block or quarantine an agent, and an audit trail that proves it. Princeton Sentinel brings those controls into one console deployed in your own Azure tenant.

How does Sentinel find shadow AI agents?

It compares the Copilot Studio agents seen in live telemetry with the agents registered in your Dataverse environment. An agent that is handling conversations but was never registered is flagged as a shadow agent, so an admin can review it and, if needed, quarantine it.

Which AI platforms does Sentinel govern?

Custom agents built in Microsoft Copilot Studio, plus Microsoft 365 Copilot usage and adoption. Agents built outside Copilot Studio aren't covered today. On the data side, Sentinel covers SharePoint, OneDrive, and Entra ID users and groups - the data and identities Copilot works with.

Does Sentinel block prompts in real time?

No. Sentinel doesn't sit inline in front of Copilot. It classifies Copilot Studio prompts after the fact for prompt-injection, jailbreak, and harmful-content risk, and gives admins the controls to act on what it finds - blocking a user from an agent or quarantining the agent.

Does Sentinel change anything in our tenant automatically?

No. Discovery, scoring, and monitoring are read-only. Revoking access, blocking a user from an agent, quarantining an agent, and disabling an agent's identity are explicit admin actions, and each one is written to an audit log you can export.

Which edition includes AI governance?

Enterprise. Sentinel is licensed per tenant: Essentials covers sync, the dashboards, Data Risk, and email reports; Professional adds permission remediation; Enterprise adds the AI governance suite - custom agents, Agent Risk, AI Safety, agent quarantine, Copilot Utilization, and AI narratives. Trials include everything.

See your AI agent inventory in the demo

We'll walk agent discovery, permission review, risk scoring, and quarantine on a tenant that looks like yours.

Request a demo