Microsoft 365 AI agent inventory
How many AI agents work for your company right now, and who hired them?
Agents ship with vendor products, get built by employees in Copilot Studio in an afternoon, and inherit whatever their builder or blueprint could reach - nobody consciously granted it. In a typical tenant starting a Copilot rollout we already find dozens of agents, and a handful of them can read hundreds of thousands of files. 1Security puts every Copilot, Entra and Foundry agent in one Microsoft 365 AI agent inventory and counts, per agent, the users, files, sites and mailboxes it can actually reach - then, once you turn write access on, revokes, blocks or grants that access from the same row, behind a review window.
- 4reach counts per agent - the users, files, sites and emails it can actually access
- 3agent sources brought into one list - Copilot Studio, Entra Agent ID, Azure AI Foundry
- 9knowledge source types resolved to real sites and mailboxes - SharePoint, OneDrive, Teams, email, websites, Graph connectors, Dataverse, people, meetings
The problem
A new workforce, onboarded by nobody
Agents are identities. Nobody hired them, nobody reviews them, and they keep their permissions until someone notices.
The person who builds an agent in Copilot Studio is thinking about a task, not about access control. The agent inherits whatever the builder could reach or whatever its blueprint declared, and the gap between "I made a small assistant" and "this thing can read the finance site" is invisible from inside the tool that created it.
Reading the manifest does not close that gap. A declared permission like Sites.Read.All names a category, not a consequence. Governance needs the list: which sites, which files, how many carry regulated data, and through which channel the access arrives. Agents arrive from several places - Copilot Studio, Entra Agent ID, Azure AI Foundry, vendor products - each with its own model and tags. 1Security brings them into one list and adds the reach.
1Security resolves it for every agent in the tenant, whether it came from Copilot Studio, Agent Builder, Entra Agent ID, Foundry or a vendor product, and keeps the blueprints in view - so one review of a template covers the fifty agents stamped out of it.
What you get
Every agent, with what it can actually read
Instances are where reach is real. Blueprints are where one review covers many agents.
The census
Every agent instance in the tenant - vendor-shipped, employee-built or bundled with a licensed product - with its creator, its blueprint and whether its account is even enabled. Usually more than anyone on the security team expected.
Reach in numbers
Users, files, sites and emails each agent can access, plus a sensitive-data flag. Sort by files in reach and a handful of agents account for most of the exposure.
File access channel
App-backed, tenant-wide, admin-consent or scoped grant - or none; email access classified the same way. The mechanism of the access, not the marketing.
Blueprints as a governed object
Publisher, verification status, declared permissions and how many cascade to instances. One unverified blueprint can be the parent of fifty agents.
Knowledge sources and capabilities
SharePoint, OneDrive, Teams, email, websites, Graph connectors, Dataverse, people and meetings - and for Azure AI Foundry agents the project connections and search knowledge bases - resolved to real sites and mailboxes, plus whether the agent can search the web, run code, generate images or take actions.
Direct, inherited, blocked
Every atomic permission with its source, and the "blocked for agents" flag - so you can see what an agent was given, what it picked up, and what you explicitly denied it.
How deep it goes
From "it has Sites.Read.All" to a list of sites
Open an agent and the Files, Sites, Users and Emails tabs show exactly what it can reach and through which channel or knowledge source.
The pre-rollout gate: filter file access tenant-wide + sensitive data within reach. Every agent on that list can quote regulated content to anyone who chats with it. In a mid-size tenant it is usually a short list - and the whole point is to see it before employees start asking questions.
The supply-chain check: in Blueprints, filter unverified publishers + with agents. These are unvetted third-party templates already running in your tenant, and the cascading-permissions count tells you how much each one hands down.
Read-and-act risk: agents with the actions capability + email access - the profile where a planted prompt does damage, because the agent ingests untrusted content and then acts on it. 1Security does not leave that to a manual review: new files and emails are scanned for injection payloads - instruction overrides, hidden text, exfiltration templates - and flagged minutes after they land, before an agent grounds on them, with Microsoft's own runtime injection verdicts arriving as instant notifications on top. The filter is for reviewing the agents that would act on what a payload says.
In practice
How a governed rollout sequences
Reach first, guardrails second, enablement third. The order matters more than the tooling.
- 01
Take the census
Open Agents. The list lands with creator, blueprint, status and reach counts - typically including agents nobody on the security team knew existed.
- 02
Sort by files in reach
Sort descending, add the sensitive-data filter. In most tenants five to ten agents hold most of the exposure; those are the first review.
- 03
Fix the data, not just the agent
Most agent over-reach is inherited oversharing. Expire the anyone links and trim the site permissions underneath, and every agent that inherits from them is constrained at once.
- 04
Grant, revoke, block - behind a review window
Set the permissions agents may not hold, revoke the grant or nested membership an over-reaching agent inherited, or grant the scoped access it actually needs. Write access is a separate consent; every action is staged as a proposal for 72 hours by default, can be routed to the owner of the resource, and lands in one audit trail. Then confirm the block shows on the row next to direct and inherited permissions - a guardrail you have not verified is a policy document.


What 1Security adds
One population, one list, with reach
Agents are created in several places. 1Security governs them as one population and adds what each one can actually read.
- Copilot, Entra Agent ID, Azure AI Foundry - discovered by walking your Azure subscriptions and projects - and third-party agents in one inventory
- Interaction volume per agent from the audit log, with imported credit reports and measured Azure token curves where agents are metered
- The concrete files and sites behind a declared permission scope, counted
- Whether sensitive data sits inside the reach of each agent
- Blueprints with publisher verification and the permissions that cascade to instances
- Direct, inherited and blocked permissions shown separately
- Knowledge sources declared per agent, resolved to real sites and mailboxes
- Agents whose creator has left or whose account is disabled but which still hold permissions
- Capability flags - web search, code interpreter, image generation, actions
- Grant, revoke and block from the same list that measured the reach - opt-in, staged behind a review window, every action logged
- Agents registered in Microsoft Agent 365 keep their control plane; their measured reach, behaviour and the unregistered ones live here
Scale
Agents multiply faster than reviews
A blueprint is copied in seconds. Governance that works per instance loses immediately, which is why the template layer is a governed object here rather than an afterthought. Reach counts come from the same permission graph that already resolves users and apps - the one that runs on tenants with 500M+ files.
- 2layers governed separately - agent instances and the blueprints they come from
- Day 1Entra Agent ID, Copilot Studio and app-layer agents on the licenses you already own; Azure AI Foundry joins through a reader-role grant on the Azure side
- 0write permissions needed for the census - blocks are a deliberate second step
Related
Where this fits
Agent reach is inherited reach. Almost every finding on this screen resolves in the file and site permissions underneath it, which is why the AI conversation and the oversharing conversation are the same conversation.
Copilot security
What Microsoft Copilot itself would read - the rollout view of the same problem.
See Copilot →App governance
The other non-human identities: OAuth apps, add-ins and managed identities, with how each reaches data.
See apps →File permissions
The layer an agent inherits from, and the fastest place to constrain it.
See files →Next to Microsoft Agent 365
How the inventory sits beside Microsoft's control plane - what a registry answers, what measured reach answers, and what each licence layer covers.
See it next to Agent 365 →The admin-center agent registry
What All agents shows, what the store shelf looks like in a real tenant, and how the registry export becomes a review.
Read the registry guide →
The 1Security platform
AI Agent Inventory is just the beginning
AI Agent Inventory is one screen of the 1Security platform for Microsoft 365. Locations, devices, apps, AI agents, logs, permissions, sensitive data, compliance, reports and actions run on the same read-only connection, on the licence you already have.
- LocationsWhere every sign-in and every action really comes from, on a map, per tenant. Breaches by location, impossible travel, allow lists per person. Works without premium sign-in logs.


- DevicesRegistered, unregistered and shadow devices, attributed from activity rather than from an agent. The device nobody enrolled still shows up.


- Apps and consentEvery app with access to the tenant, what it can reach, who consented, and whether it still needs to.


- Activity logs and forensicsUp to three years of searchable activity on a standard licence. What was opened, downloaded, shared, by whom, from where, on which device.


- Anomalies and detectionsSpikes in sharing, downloads, sign-ins, consent and agent activity, scored against each tenant’s own pattern, with curated response actions.


- Permissions and sharingWho can reach what, through which group, link or guest, and whether anyone meant it. The sharing map every audit asks for.


- Microsoft 365 Alerts ToolOne alert model across policies, activity and anomalies - with the evidence attached and the alert threshold in your hands.
- Microsoft 365 Monitoring ToolReal-time monitoring that catches a mass download while it is unfolding - across users, files, apps, AI agents and devices.
- Copilot Security ToolEvery AI agent's real blast radius - quantified before rollout, watched and constrained after.
- Shadow Device DetectionEvery device touching your data, reconstructed from real activity - including the machines that never enrolled anywhere.
- Impossible Travel DetectionLocations as identities, not IPs: country, city and network for every action, first-seen origins, travel timelines with verdicts - no E5, no P2.
- Conditional Access MonitoringDeclared Conditional Access compared against observed sign-ins: coverage verdicts, ungoverned traffic share, and the ranges your offices outgrew.
- Office 365 Access ManagementEffective access, resolved - who can reach what, through which grant, and one place to take it away.
- SharePoint Governance ToolEvery site ranked by exposure and abandonment - links, guests and Copilot access under control.
- Office 365 User ProvisioningLifecycle access from day one to offboarding - reach, dormant accounts and license reclaim in one flow.
- Microsoft 365 Audit ToolUp to three years of searchable activity history - every action tied to an actor, device and location.
- Microsoft 365 Inventory ToolEight live inventories on one permission graph - including the sites, devices and AI agents that were never registered anywhere.
- Office 365 ReportingAnswers, not exports - regulator-grade numbers, saved views and a read-only API for your SOC.
- Microsoft 365 File PermissionsWho has access to any file, and why - people, links, apps and AI agents on one map, with the fix in the same place.
- Sensitive Data DiscoveryWhere regulated data lives and everything that can reach it - two detection engines, no premium licence required.
- Office 365 Email MonitoringOutbound sensitive content, delegated-send abuse and whole-thread evidence that survives deletion.
- Purview Label CoverageWhat your sensitivity labels actually cover, and which of them protect anything at all.
- AI Data RestrictionDeclare a data type off-limits to AI and every agent, app and user still reaching it becomes a live violation - recorded for the auditor.
- AI Spend TrackingInteractions, Copilot credits and Azure tokens in one ledger per agent - with the spend column next to what each agent can reach.
- Microsoft 365 User Access ReviewEvery identity with what it can reach, what it did, and whether it should still exist - including SharePoint-only guests that live outside the directory.
- Group Access AnalysisWhat membership in a group actually unlocks - nesting resolved, external members surfaced, blast radius counted.
- App and OAuth Consent GovernanceEvery app with a foothold in the tenant: how it got in, who let it in, and what it can actually read.
- Remediation AutomationFindings become fixes at tenant scale, counted before you enable them and staged behind a review window.
- Microsoft 365 Licence OptimizationThe seats you pay for and nobody uses - unassigned units, licences on guests, and dormant accounts still billing.
- Microsoft 365 Anomaly DetectionEvery policy, user, app and agent measured against its own baseline - with an alert line you move and see the answer to at once.
- Office 365 Sign-in LocationsEvery place your tenant is reached from, with your Conditional Access configuration laid over observed reality.
- Microsoft 365 Unified Audit LogWho did what, when, from where and on which device - retained up to three years on a standard licence.
- SharePoint Site InventoryEvery site, channel site and OneDrive in one list, ranked by external reach, abandonment and what Copilot can read.
- Microsoft 365 Activity AnalyticsThe extremes rather than the averages - a mass download inside the hour, and everything nobody has touched in a year.
- Microsoft 365 Trend TrackingAny number in the tenant followed over time, with the change and the deviation attached - and alerting on the same object.
- Microsoft 365 Compliance MonitoringReadiness for eleven regulations measured from live tenant data - with weekly snapshots and a dated evidence pack per framework.
- Microsoft 365 Security ScoreMicrosoft Secure Score beside the score measured from what actually happens - one worklist ranked by points still on the table.
- External Domain MonitoringEvery organisation your tenant exchanges data with, resolved to a real company - with drift flags and an enforceable verdict per domain.
- All featuresExplore everything
FAQ
Questions teams ask first
Which agents are covered?
Copilot agents built in Copilot Studio or Agent Builder, Entra Agent ID agents - including the earlier generation that exist as tagged service principals - and vendor-shipped agents that arrive with a licensed product. Azure AI Foundry agents have their own discovery path: 1Security walks your Azure subscriptions and Foundry projects, matches each agent to its Entra identity, and brings its instructions, tools and knowledge sources along. The inventory covers them wherever they were created.
Do we need a special license?
No. Entra Agent ID, Copilot Studio, declarative Copilot and app-layer agents arrive through the standard read-only connection, on the licenses you already own. Azure AI Foundry goes deeper with one extra step on the Azure side - two reader roles on the subscriptions and projects - which is a role assignment, not a license. Where a specific control runs through Microsoft Agent 365, such as blocking a declarative agent's store package, 1Security follows the license you hold, and the inventory says plainly which layer each agent came from.
What is a blueprint?
The template an agent instance is stamped out of. Reviewing it once - publisher, verification, declared permissions, how many permissions cascade - covers every instance created from it, which is the only way agent governance keeps up with agent creation.
Can I actually stop an agent from reading something?
Yes, once write access is turned on - a separate consent the census never needs. Permissions can be blocked for agents, and the block is verifiable on the agent row next to its direct and inherited permissions. You can also declare a sensitive information type off-limits to AI: every agent that can still reach it is listed as a live violation, and the declaration is recorded with who made it and when. The more durable fix is removing the underlying oversharing - revoke the grant or nested membership, close the site's external sharing, expire the anyone-links - because that constrains every agent at once. Each of those runs as a staged proposal inside a review window, optionally routed to the resource owner, and is recorded in one audit trail.
How does this relate to Microsoft Agent 365?
Agent 365 is the control plane for the agents you register - identity, lifecycle, policy. 1Security is the measurement layer under it: what each agent can actually reach resolved to files, sites and mailboxes, how it behaves against its own baseline, who runs it - and the agents nobody registered, which a registry cannot show by definition. They run together.
What happens to agents whose creator has left?
They keep working. Orphaned agents are their own filter, and the right treatment is the one you would give an orphaned service account: identify an owner or decommission it - behind the same review window as every other action.
Take the census before the rollout.
Connect read-only and every agent in the tenant arrives with its reach attached the same day - including the ones built in an afternoon.
Or find out what an agent can read from a generated answer.