Identity Mesh — November 10, 2026
Release Date: November 10, 2026
Release Type: Mixed (New Features, Improvements, Bug Fixes, Security Fixes)
Overview
This release puts you in control of the connectors and on-prem agents you no longer want, and attaches accounts that until now had nothing to match on:
- Remove a connector you no longer use — take a connector out of service yourself, and keep everything it collected.
- Retire an on-prem agent for good — end an agent's enrollment from Mesh, and stop an agent retiring itself over a passing connection problem.
- Accounts known only by a login now link to the person — the accounts your endpoint tools report for AI agents join the rest of that person's identity.
- Open and resolved risks are told apart — the identity page counts what is still open.
It also adds sorting and filtering to the Algorithm Posture Matrix drill-down, renames the AI gateway settings, corrects a phishable-credential finding raised in error on Okta passwords, fixes five defects, and updates third-party components for published vulnerabilities.
New features
Remove a connector you no longer use
A connector you created by mistake, or one for a system you have retired, could be disabled but not taken out of your list. You can now remove it yourself.
What you get:
-
Remove connector in the connector's actions menu, and in the Okta, AWS and Kubernetes edit panels.
-
What you have already collected stays. Identities, accounts and findings collected by that connector remain in Mesh, so your history and your reports stay accurate.
-
Nothing changes in the source system. Removing a connector stops Mesh collecting from it; it does not touch the system it was reading.
-
A reason is required, and it is recorded in the audit trail with who removed the connector and when.
-
Sync progress is cleared. If you add the same connector again later, collection starts fresh rather than carrying on from where the old one stopped.
-
The connector leaves your connector list. Restoring it, or deleting the data it collected, is a support request.
Why it matters: duplicate connectors left over from onboarding, and connectors for systems you no longer run, used to stay on the list unless someone disabled them one at a time. Removing one is now a deliberate, recorded action that costs you none of the history you have built.
Retire an on-prem agent for good
The on-prem agent now tells a genuine retirement apart from a connection it cannot currently make, and treats the two very differently.
What you get:
-
Retiring an agent from Mesh is final. The agent erases its credentials and its stored configuration, and will not start again on that machine. Bringing that machine back into service means enrolling it afresh.
-
An agent no longer retires itself over a passing problem. Where Mesh cannot confirm that an agent was retired, the agent keeps trying and keeps its data, so a refusal sent in error costs you nothing.
-
The agent list says when a connection is being refused. A Connection refused badge gives the reason and the next step — an older install of the same agent still running, an agent identity used from another machine, or an agent with no machine on record.
Why it matters: an agent that retired itself because of a certificate renewal or a brief outage had to be enrolled again by hand, and the screens gave no reason why it had stopped. Retirement is now something you decide, and everything else is something you can see and act on.
Accounts known only by a login now link to the person
Endpoint tools report the user an AI agent runs as by login alone — sam, with no email address and usually no name. Those accounts had nothing to match on and stayed separate from the person they belong to. They now link.
What you get:
-
An account whose sign-in name is a bare login, with no
@domainpart, is linked to the one person in your tenant whose email address begins with that login —samis matched tosam@example.com. -
Mesh links only where the match is unambiguous. Exactly one person's email address may match; shared and generic logins such as
admin,supportorsvcare never matched; and anything already known that disagrees — a different first name, last name or employee ID — stops the link. Two bare logins are never linked to each other. -
Accounts you have already collected are linked on the next correlation run. You do not need to re-sync a connector.
Why it matters: CrowdStrike and Cortex XDR identify an AI agent's user this way, so the agents they found appeared with no owner and raised an unowned-identity risk even when that person was already in Mesh under their directory account. One identity now carries both.
Open and resolved risks are told apart
Every risk on an identity now records whether it has been resolved, and the identity page counts accordingly.
What you get:
-
Open risks on an identity and its accounts are counted on their own. A risk that has since been resolved no longer adds to the total.
-
The Risks tile on the identity header opens the risk findings for that identity, so the count and the list it refers to are one click apart.
Why it matters: the count on the identity page included risks that had already been resolved, so it did not match the findings list underneath it and overstated what was left to do. Expect this count to fall for identities with a history of resolved risks — nothing has been removed, it is being counted correctly.
Improvements
Sort and filter the Algorithm Posture Matrix drill-down
Opening a grade tile in the Algorithm Posture Matrix now gives you a list you can order and narrow rather than scroll: sort by name, type or algorithm, and filter by type or by name. The result count covers the whole filtered set, not just the page you are looking at.
The AI gateway settings have clearer names
In Settings, Agent policies is now Runtime AI Gateway. Inside it, AI policy is now Runtime Agent Policies and Agent access is now Target Access, and the signal bands sit beside them on their own tab. Your policies, their version history and the version you have enabled are unchanged — only the names on screen have changed.
Okta passwords are no longer reported as phishable
Mesh does not count a password as a phishable credential when the identity provider requires it of everyone. Microsoft Entra marks its password that way and the Okta connector did not, so Okta passwords were reported as phishable credentials in error. The Okta connector now marks it correctly.
Existing connectors pick up this correction after you open the connector, select Edit, then Save. The credentials it covers are then re-checked and the findings raised in error clear. A re-sync is not needed.
Clearer summary cards in the agentic AI inventory
The summary cards above the AI agent list are now grouped under How we know and Agent policy, using the same words as the table column and the Filters menu. "Suspected" and "Suspected shadow AI" previously sat side by side with nothing to say they answered different questions.
Audit log filters that can return something
A second contains, does not contain, in or not in filter on the same field now joins the first rather than combining with it to match nothing. A combination that cannot match, a time filter outside the dates you have selected, and a filter identical to one already applied are refused with a note naming the filters that conflict, instead of quietly returning an empty table.
Dark-web breach checks keep up
A repeat breach check for an identity that had not changed could spend a fresh external lookup instead of reusing the last result, and a backlog of pending checks built up behind it. Breach results now arrive without that backlog.
Nightly analytics rides out network interruptions
A nightly export interrupted part-way is now sent again automatically instead of failing the run, and without duplicating what had already been written.
Newly collected data appears sooner
Collection repeats less work on the largest tables, so identities and machines you have just collected show up sooner on large tenants.
Bug fixes
-
Okta identities with application assignments — a partial record from Okta replaced the person's name and email address with blanks, so those identities showed an empty profile and never linked to their other accounts.
-
Identity risk score — an identity's risk stayed at zero when its credential risks were collected before the identity itself.
-
Correlated identities — Open correlated peers on the identity page opened an empty drill-down.
-
Nightly compliance data — the nightly compliance update could fail to complete on large tenants, leaving your compliance data unrefreshed and then rejected as out of date.
-
Daily risk-trend snapshot — tenants with no risk groups were retried every five minutes instead of being skipped.
Security fixes
- Third-party components used by Mesh services, connectors and the on-prem agent are updated for published vulnerabilities, including CVE-2026-78661, CVE-2026-116689 and CVE-2026-49272.