Fix it now
These come from the pass-through authentication path, not from the user’s account. Microsoft documents AADSTS80001 as the Authentication Agent being unable to connect to Active Directory, and AADSTS80002 as the password validation request timing out. AADSTS50144 is the exception: that one really is the account, because the on-premises password has expired.
Get-Service | Where-Object {$_.DisplayName -like '*Authentication Agent*'}
nltest /dsgetdc:contoso.local /writable
Invoke-ADSyncDiagnostics -PasswordSync
- Restart both services on each agent server, the updater first and then the agent, and watch the agent’s own log as it comes back.
- Confirm the domain controller lookup returns a writable controller in the expected site. Agent servers must be members of the same forest as the users whose passwords are validated.
- In the Microsoft Entra admin center, open the Entra Connect area and check the pass-through authentication agent list. Every agent you installed should show as active.
- Test a sign-in from a private browser window with a known-good account. If it works for some users and not others, go to the expired-password cause below.
- Read the diagnostics output for password hash synchronisation, which is your fallback if every agent goes down.
Microsoft recommends a minimum of 3 Authentication Agents running in production, with a system limit of 40 per tenant. Two is not the documented figure.
If sign-ins recover and the agent list is healthy, you are done. If they come and go under load, the next section separates the connection failure from the timeout, which need different fixes.
Why it happens
Pass-through authentication does not send passwords to the cloud to be checked. You install a lightweight agent on a domain-joined server; it opens outbound connections to Microsoft and waits. When a user signs in, Entra ID encrypts the password and queues it, an agent collects it, decrypts it and validates it against a domain controller, and the verdict – success, failure, expired, locked out or disabled – goes back the same way.
Every part of that chain is yours except the queue, and the requirements are documented. Agents make outbound connections only – there are no inbound rules to open – and the server must sit in the same Active Directory forest as the users being validated, must not be Server Core, and should be treated as a Tier 0 system and hardened as you would a domain controller. It still needs what any domain member needs to reach a controller: Kerberos, LDAP, the RPC endpoint mapper with its dynamic range, SMB and DNS. Block any of those and the agent runs perfectly while validating nothing.
That is exactly the shape of AADSTS80001, which Microsoft documents as the agent being unable to connect to Active Directory. AADSTS80002 and AADSTS80014 are a different failure in similar clothing: timing. The request was collected by an agent and no answer came back inside the window the service allows, which points at agent count, controller health or something inspecting the agent process rather than at connectivity.
The agent service is stopped, or the server hosting it is gone
You have this one if AADSTS80001 for everyone, and the agent list shows no active agents or fewer than you installed.
- Check the service state on each agent server, then restart both the agent and its updater.
- Confirm the agent reappears as active in the portal before changing anything else.
- Get to at least three agents on separate servers, which is the documented production minimum.
Agents need not live on domain controllers. Spreading them across member servers reduces the blast radius of a reboot.
The agent cannot reach a domain controller
You have this one if The service is running and shows as active, and AADSTS80001 continues while the agent log records failures locating or binding to a directory server.
- Run the writable domain controller lookup and confirm it returns a controller in the expected site.
- Check the server’s DNS points only at controllers hosting the AD zone; a public resolver there breaks discovery in a way that looks like a network outage.
- Confirm 88, 389, 135, 445, 53 and the high RPC range are open towards your domain controllers.
- An agent in a site with no local controller sends every validation across a WAN link. Move it, or add a controller.
Validation is timing out rather than failing
You have this one if AADSTS80002 or AADSTS80014, intermittently, and worse at the start of the working day.
- Add agents. This is the single most effective fix and there is no licence cost to running more of them, up to the documented limit of 40.
- Exclude the agent’s program directory and processes from on-access antivirus scanning, through your change process.
- Check domain controller health: high LSASS CPU or a saturated link produces exactly this pattern.
- Confirm a backup window is not overlapping the morning sign-in peak on the agent servers.
Timeouts often come from one agent that is up but sick. Restarting a single unhealthy agent frequently clears a fault that looks tenant-wide.
The on-premises password has expired
You have this one if AADSTS50144 for one user while everybody else signs in normally.
- Check the account for an expired password, a lockout or a disabled state.
- Reset the password on-premises, or have the user change it on a domain-joined device.
- To let users change an expired password during cloud sign-in, enable that setting for pass-through authentication before you need it.
Password hash sync, the fallback, is stale or broken
You have this one if Event ID 611 in the Application log on the Entra Connect server, from the Directory Synchronization source.
- Run the password sync diagnostics and read the failing step it reports.
- Confirm the AD connector account still holds Replicate Directory Changes and Replicate Directory Changes All at the domain root.
- Check for the heartbeat, Event ID 654, which should appear every 30 minutes; its absence is the clearest sign the feature is not running.
- If hashes are stale rather than absent, re-prime them with Microsoft’s documented force-full-password-sync procedure – and run that script only once.
Keeping password hash sync enabled alongside pass-through authentication is not a contradiction. It is the documented way to keep signing in when every agent is down.
Full reference
Match the symptom to the layer
| What you observe | Where the fault actually is |
|---|---|
| Every user fails at once | All agents are down, or their servers cannot reach a domain controller |
| Failures come and go under load | Too few agents, or one unhealthy agent still taking work |
| Only one site or domain fails | Agents cannot reach the controllers authoritative for those users |
| One user fails, everyone else is fine | Expired password, locked-out or disabled account, or never synchronised |
| Sign-in works but with the old password | Password hash sync is serving a stale hash and the tenant fell back to it |
What the agent servers actually need
| Direction | Port | Purpose |
|---|---|---|
| Outbound to Microsoft | TCP 443 | All communication with the service |
| Outbound to Microsoft | TCP 80 | Certificate revocation lists during TLS validation |
| Outbound to Microsoft | TCP 8080, optional | Status reporting every 10 minutes if 443 is unavailable; never used for sign-ins |
| Inbound | None | Agents make outbound connections only |
| To domain controllers | 88, 389, 135, 445, 53 and the high RPC range | The ports any domain member needs to authenticate against a controller |
The URL list Microsoft publishes for the agents includes the service bus and application proxy endpoints as well as the sign-in endpoints, plus a set of certificate validation hosts on port 80. Allow the documented list rather than a subset you assembled from a packet capture, because the failure mode when one is missing is intermittent rather than obvious.
Password hash synchronisation events worth knowing
| Event ID | What it records |
|---|---|
| 611 | An error during password hash sync for a domain |
| 652 | An error during a password hash sync operation |
| 654 | The end of a password hash sync ping – the heartbeat, expected every 30 minutes |
| 655 | An error during a password hash sync ping |
| 617 and 618 | A full password hash sync started, for a forest and for a domain |
| 622 and 623 | A full password hash sync completed, for a domain and for a forest |
Those are logged in the Application log under the Directory Synchronization source on the Entra Connect server. The one to check first is 654: if the heartbeat has not appeared in the last three hours, the feature is not running at all and nothing else you read will make sense.
Diagnostics rather than guesswork
Invoke-ADSyncDiagnostics -PasswordSyncchecks the feature end to end for the whole tenant.- The same cmdlet with
-ADConnectorNameand-DistinguishedNamenarrows it to one user object. - The Entra Connect wizard’s Troubleshoot task offers the same checks with a prompt for whether the problem affects everyone or one account.
- The checks it runs are worth knowing: the feature is enabled, the server is not in staging mode, the heartbeat has appeared in the last three hours, the AD DS account’s credentials and permissions are valid, and the domain is reachable.
- Staging mode disables password synchronisation entirely, which is the answer to a surprising share of these tickets.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
AADSTS80001 |
The Authentication Agent is unable to connect to Active Directory; agent servers must be members of the same AD forest as the users whose passwords are validated and must be able to connect to it | Microsoft Learn |
AADSTS80002 |
The password validation request timed out; Active Directory must be available and responding to requests from the agents | Microsoft Learn |
AADSTS80014 |
The validation request responded after the maximum elapsed time was exceeded | Microsoft Learn |
AADSTS50144 |
The user’s Active Directory password has expired; generate a new password or have the user reset it | Microsoft Learn |
Event ID 611 |
An error during password hash sync for a domain, logged in the Application log by the Directory Synchronization source | Microsoft Learn |
Confirm the fix worked
- Every agent you installed shows as active in the pass-through authentication view, and there are at least three of them.
- A test user signs in and the sign-in record shows pass-through authentication rather than a fallback.
- Stopping the agent service on one server deliberately leaves sign-ins working through the others.
- Event ID 654 has appeared on the Entra Connect server within the last 30 minutes, and no new 611 entries follow a full cycle.
Questions people ask about this
Does fixing this cost anything?
No. The authentication agents are part of the hybrid identity features and carry no separate charge, so you can run as many as you need up to the documented limit of 40 per tenant. The fix is service restarts, firewall rules and more agents.
How many agents should we run?
Microsoft recommends a minimum of 3 in production. Spread them across separate servers so a single patch window never leaves the tenant unable to validate a password.
Do the agents need inbound firewall rules?
No. They make outbound connections only – 443 for service traffic, 80 for certificate revocation checks, and optionally 8080 for status reporting. Anything asking you to open an inbound port is not this.
Should we keep password hash sync switched on as well?
Yes, unless policy forbids it. It is the supported fallback: if every agent goes offline, sign-ins continue against the synchronised hashes. The trade-off is that a hash of the password leaves your network, which some organisations will not accept.
Why does one user get AADSTS50144 when nobody else has a problem?
That code is not about the agents. It reports that the account’s on-premises Active Directory password has expired. Reset it, or enable the setting that lets users change an expired password during cloud sign-in.
