Microsoft 365 grew faster than anyone could govern it. Sites nobody owns, anonymous links nobody remembers creating, and now Copilot agents reading all of it. Princeton Sentinel scores every SharePoint site for exposure, tracks every AI agent's reach, and gives your admins a remediation queue instead of a hunch - deployed inside your own tenant.
Microsoft Entra ID sign-in · Read-only by default · Your data never leaves your tenant
Built on the Microsoft stack you already run
Nothing to install on endpoints. No copy of your content shipped to a vendor cloud. Sentinel runs where your Microsoft 365 data already lives.
Register Sentinel in your Entra ID with read-only Microsoft Graph permissions and deploy it into your own Azure subscription. Because discovery is read-only, the first run cannot change anything - which is usually what gets it approved quickly.
A scheduled worker syncs SharePoint, OneDrive, sharing and Copilot telemetry into your database, then scores every site and every agent for exposure using weighted signals rather than a single alarming number.
Dashboards surface the highest-risk sites and agents with plain-language summaries. Admins revoke links, quarantine agents and export audit logs in one click - or hand the queue to the department owners who actually know the content.
Over-shared data and ungoverned AI agents are the same risk seen from two angles. Each card links through to the detail on the Princeton Sentinel site.
Every SharePoint site gets an exposure score built from real signals - anonymous links, external access, dormancy - with an AI-written summary explaining why, and one-click revoke to fix it.
Explore data risk scoringCopilot Studio agents scored for reach, tool access and credential hygiene, with live telemetry, safety-incident tracking and a quarantine control for the ones that should not have shipped.
Explore AI agent securityAnonymous, organization-wide and direct links, guests and external users, mapped across the whole estate. It is the exposure view the SharePoint admin centre never gave you.
Tour the sharing mapDiscovery and scoring need only read permissions, so Sentinel cannot alter your environment on its own. Every revoke, block and quarantine is an administrator's decision, logged for audit.
See the architectureAssign sites to data stewards who see only their own scope. Security keeps oversight and the audit trail, while the people who know whether a document should be public do the remediation.
Take the platform tourExportable audit logs, historical posture trends and per-site evidence - so "who could see this, and when" has an answer that does not involve three people and a weekend.
Read the technical FAQA link created for one supplier in 2021 does not expire because the project did. Multiply that across every site, every team and every departing employee, and "who can reach this content" stops being a question anyone can answer from the admin centre.
How data risk scoring worksCopilot Studio put agent-building in the hands of the business, which is the point of it - and also the problem. An agent inherits reach, holds credentials, and answers questions all day without anyone reviewing what it is grounded on.
How agent security worksThe full technical list lives in the Princeton Sentinel FAQ.
Sentinel deploys into your own Azure subscription and reads your Microsoft 365 estate through Microsoft Graph. Findings are written to your own database. No copy of your content is shipped to a vendor cloud, and there is nothing multi-tenant about the deployment.
It registers in your Entra ID with read-only Microsoft Graph permissions by default, so discovery and scoring cannot alter your environment. Remediation actions such as revoking a sharing link or quarantining an agent are always initiated by an administrator, never automatically.
Each SharePoint site is scored on five weighted factors drawn from live signals, including anonymous sharing links, external and guest access, and dormancy. The dashboard shows the contributing factors in plain language, so a score can be explained to the business rather than simply asserted.
No. Sentinel is built on Microsoft Graph rather than on E5-only compliance features, which is much of the point: it gives organizations on lower licence tiers a view of data exposure and AI agent risk they would otherwise have to buy up to get.
Yes. Sites can be assigned to data stewards who see and remediate only their own scope. Central security keeps oversight and the audit trail, while the people who actually know whether a document should be shared do the cleanup.
Princeton Sentinel has its own site with the full platform tour, the scoring methodology, architecture and data-residency detail, and real product screenshots.
A thirty-minute walkthrough against a demo estate, and a straight conversation about what a read-only connection to yours would surface first.