Identity Mesh — November 3, 2026
Release Date: November 3, 2026
Release Type: Mixed (New Features, Improvements, Bug Fixes)
Overview
This release puts the AI agents in your directory into the identity list beside everything else, and makes it quicker to find one identity among many:
- AI agents and more non-human identities from Microsoft Entra — agent identities, the blueprint behind each one, and five further kinds of non-human identity.
- Find one user, or everyone from one connector — search across the attributes you know a person by, and filter the identity list to a single connector.
- SailPoint Identity Security Cloud — a cloud-hosted SailPoint connector you pick from the list.
- On-prem agents settings — the on-prem area becomes a set of pages, and connectors are added through the standard connector screens.
It also makes the nightly analytics and enrichment work faster, and fixes fourteen defects.
New features
AI agents and more non-human identities from Microsoft Entra
The Azure connector now collects the identities Microsoft Entra creates for AI agents, along with the blueprint each agent is built from, and fills in the kinds of non-human identity it previously skipped.
What you get:
-
Agent identities, with their blueprint. Each agent identity is collected together with the blueprint that governs it, and the two stay distinguishable — you can see both the agent and the template it came from.
-
Five more kinds of non-human identity — OAuth applications, workload identities, managed identities, CI/CD runners and service accounts. The non-human identity list now reflects what is in the directory rather than a subset of it.
-
Agent sign-in activity, where Microsoft Defender is licensed. Mesh collects agent sign-in activity through Defender and marks a service-principal sign-in as agent activity when the principal signing in is an agent identity, carrying that agent's blueprint. Sign-ins by ordinary service principals are not marked. A tenant without the required Defender license, or with no qualifying activity, completes collection normally rather than reporting a failure.
Why it matters: AI agents authenticate and act like any other identity, but they arrive through a different part of the directory. Collecting them alongside your service accounts and workload identities means one list answers what non-human identities exist, what each is for, and which of them are agents.
Find one user, or everyone from one connector
Identity search now answers the two questions that come up most during onboarding and connector troubleshooting: what does Mesh know about one person, and did a connector bring everyone in?
What you get:
-
Search by what you know. Match on username, email, display name, employee ID or the account ID in the source system. Matching is case-insensitive and works on partial values.
-
A result that shows the correlated identity, and which connector each matched account came from.
-
Filter the identity list to one connector. The filter selects a configured connector, not a provider type, so two tenants of the same provider are distinguishable. It combines with the existing status, risk and identity-type filters and with search, and shows the result count.
-
Accounts that correlated to nobody are included and marked as such, so a connector's orphans are visible rather than hidden.
-
Export and query the same result. Filtered results export to CSV, and the same query is available through the analytics API.
Why it matters: without these, answering "is this person in Mesh" or "did the sync bring everyone in" meant an export or a long scroll.
SailPoint Identity Security Cloud
A SailPoint Identity Security Cloud connector is now one you pick from Add connector, for cloud-hosted SailPoint tenants. It joins the SailPoint IdentityIQ connector released on October 27, which collects from an on-premises deployment through an on-prem agent.
On-prem agents settings
The on-prem agents area is now a set of pages rather than overlays stacked inside one another, so a step is a place you can navigate to and return to. Adding a connector to an agent uses the same connector screens as everywhere else in Mesh, instead of a separate flow with its own layout.
Improvements
Connectors and syncs use the same words
The Connectors and Data source syncs screens described the same things differently — one screen's "source" was another's "connector". Both now use Connector for the instance you configured and name the vendor separately, so a value you read on one screen means the same on the other. The Data source syncs CSV export follows the same naming: its Source and Connector Type columns are now Connector and Provider, so a script that reads the export by column name needs updating.
AI agents found by a website lookup now show an owner
An AI tool seen only as network traffic to its service is labeled Suspected shadow AI. Those detections carry no user of their own, so they appeared with no owner and raised an unowned-identity risk. Where an endpoint tool saw exactly one person on the same computer, that person is now shown as the owner, marked "same computer", and the unowned-identity risk is cleared. On a shared computer the agent keeps no owner and lists the possible users.
Faster certificate-to-machine linking from ServiceNow CMDB
The ServiceNow CMDB connector links each certificate to the computer whose fully qualified domain name, or failing that whose host name, matches the certificate name. Collection now makes far fewer calls to your ServiceNow instance and finishes sooner. The links it finds are the same as before.
A short network interruption during this step no longer marks the sync as failed: the links already made are kept and the rest are retried on the next run. If the next run hits the same interruption, the sync is reported as failed.
Existing connectors pick up this improvement after you open the connector, select Edit, then Save. A re-sync is not needed.
Nightly reconciliation finishes faster
The nightly job that reconciles identities and removes orphaned records now completes reliably and finishes much faster, including on large tenants.
Faster, more efficient AI enrichment
Enrichment no longer asks the AI about the same identity more than once, and network endpoints that cannot be enriched are skipped rather than retried. Together these make each nightly run faster.
A connector runs one sync at a time
A long-running connector sync could be started a second time while the first was still running. A connector now runs one sync at a time, however long the sync takes.
Bug fixes
-
Widget chat history — per-widget chat history was lost when navigating between views.
-
Widget chat answers — chat in a widget could not link an identity to its assets, and could answer about the wrong user.
-
Audit log filter — the "contains" filter returned no matches.
-
Connector health — a connector that collected all its identity data showed as "Unknown" when only its activity data, such as sign-ins, could not be collected; it now shows Healthy, and Sync status still reports the data that was not collected.
-
Connector data loss — when a sync failed to deliver its collected records, those records were lost and the sync was still recorded as successful. A failed delivery is now reported as a failed sync.
-
Okta first load — a duplicate record in the Okta data could stop a first load from completing.
-
Okta connector — four of the setup wizard's checks timed out on Okta orgs with many applications, and a sync could be reported as failed because of applications Mesh excludes from collection.
-
Search timeouts — searching an uncommon name in the drill-down and posture matrix timed out on large tenants. Search now starts from the third character.
-
Impacted users — the drill-down was empty for the "agent with no owner" risk, which was also labeled as a different risk.
-
Dark-web breach checks — a breach check that ran long could leave its risk results unconfirmed; it is now run again automatically.
-
ServiceNow in government clouds — the connector could not use a FedRAMP ServiceNow endpoint. It can now.
-
Compliance export — the compliance snapshot export failed on every read.
-
Compliance data — a run that found no compliance frameworks replaced your stored compliance data with nothing and reported success. Stored compliance data is now kept.
-
Nightly analytics export — the nightly export could fail to complete, and one set of inventory data was missing from it.