Fix it now
0x80090326 is SEC_E_ILLEGAL_MESSAGE, which Microsoft documents as the Schannel error code returned for a TLS alert – and as the default for any alert it does not map separately. Your credentials were never reached. The TLS handshake underneath the connection was rejected, so look at what the two machines can agree on before you touch any policy.
Get-TlsCipherSuite | Select-Object -ExpandProperty Name
Get-HotFix | Sort-Object InstalledOn -Descending | Select-Object -First 5
- Compare the cipher suite lists. If the two machines have nothing in common the handshake cannot complete, and that is the most direct reading of this code.
- Check whether a hardening template, a security baseline or an old registry-based Schannel configuration has disabled protocol versions on one side. Anything that changed the enabled protocols since the connection last worked is your first suspect.
- On the host, confirm the Remote Desktop listener has a usable certificate: not expired, with a private key, and trusted by the client if the client has been configured to require it.
- Check whether anything on the path inspects or rewrites TLS. A device substituting its own certificate, or one that does not support the versions in use, produces a rejected handshake and nothing more informative.
- Bring the machine that is behind up to the current cumulative update and reboot it. A stale security stack is a common reason two machines have nothing in common.
The widely repeated advice to set Encryption Oracle Remediation to Vulnerable is aimed at a different failure. Microsoft documents that one as reporting that the function requested is not supported, not this code.
If the connection completes you can stop here. If it does not, the next section explains what Schannel is telling you and which of the two authentication stories you are actually in.
Why it happens
A Remote Desktop connection with Network Level Authentication runs its credential exchange inside TLS. Schannel is the Windows component that negotiates that TLS session, and when the handshake fails it converts the alert it received into an error code. Microsoft publishes that mapping, and 0x80090326, SEC_E_ILLEGAL_MESSAGE, is the busiest entry in it: it is the code for unexpected_message, record_overflow, handshake_failure, illegal_parameter, decode_error, export_restriction, no_renegotiation, unsupported_extension and unknown_psk_identity, and it is the default for any alert not listed separately.
That is why it tells you so little on its own and so much once you know what it is. It is not a statement about your password, your account or the host’s health. It is Schannel saying the TLS session could not be established, and the causes are the ordinary causes of a failed TLS handshake: no protocol version in common, no cipher suite in common, a certificate the other side would not accept, or something on the path interfering with the exchange.
The CredSSP story belongs here as a branch rather than as the whole article. Microsoft’s own article for that update documents its symptom as an authentication error saying the function requested is not supported, with later updates adding that this could be due to CredSSP encryption oracle remediation. That is a different code. Encryption Oracle Remediation is worth checking on a machine carrying an old workaround, but reaching for it first will send you into Group Policy over a cipher suite.
The companions come from the same security layer and two of them are documented in the same Schannel mapping. 0x80090304, SEC_E_INTERNAL_ERROR, is the code for the TLS internal_error alert. 0x80090330, SEC_E_DECRYPT_FAILURE, is the code for the decryption_failed and decrypt_error alerts. 0x8009030D and 0x80090329 are not in that mapping and Microsoft publishes no meaning for them in a reachable table, so treat them as statuses from the same package rather than as diagnoses.
The two machines have no protocol version or cipher suite in common
You have this one if One end has been hardened, or has a security baseline applied, and the failure started after that rather than after a Windows update.
- Compare
Get-TlsCipherSuiteoutput on both machines and look for an intersection. - Review any hardening template or baseline applied to either machine, and the Schannel protocol settings it set.
- Re-enable a protocol version or suite that both can support, then plan the removal properly rather than leaving one end unable to speak to the other.
The listener certificate on the host is unusable
You have this one if Every client fails against this host, and other hosts on the same network are fine.
- On the host, check the certificate the Remote Desktop listener is using: in date, with a private key present.
- Where the client is configured to require a trusted certificate, confirm the issuing chain is trusted on the client.
- Replace or reissue the certificate, then restart the Remote Desktop Services service and retest.
Something on the path is inspecting or rewriting TLS
You have this one if It fails through one network path and works on another, or only through a particular appliance or proxy.
- Test the same client and host on a path with no inspection device in it.
- Where inspection is deliberate, exclude the Remote Desktop traffic from it rather than weakening either endpoint.
- Check the appliance’s own logs. It will normally record the handshake it rejected and why.
One machine’s security stack is well behind the other
You have this one if The failing pair is one current machine and one that has not been updated for a long time.
- Install the current cumulative update on the machine that is behind, and reboot it.
- For a thin client or appliance, check with its vendor for firmware that includes the current protocol support.
- Where the machine is past end of support it will not receive the update. Replace it rather than lowering the other end.
An old CredSSP workaround is still in place
You have this one if The documented CredSSP symptom is a message saying the function requested is not supported, and a machine carries an Encryption Oracle Remediation setting nobody remembers applying.
- Check the value:
reg query "HKLM\Software\Microsoft\Windows\CurrentVersion\Policies\System\CredSSP\Parameters" /v AllowEncryptionOracle. - If a Group Policy object is setting it, remove the setting there rather than deleting the value locally.
- Return both ends to Force Updated Clients, run
gpupdate /force, and retest.
Search your Group Policy objects for Encryption Oracle Remediation before concluding a machine is unmanaged. These settings were applied widely as an emergency measure and were often never reversed.
Full reference
The five statuses, and what is published
| Code | What Microsoft publishes |
|---|---|
0x80090326 |
SEC_E_ILLEGAL_MESSAGE: the Schannel error code for a TLS alert, including unexpected_message, handshake_failure, illegal_parameter and decode_error, and the default for any alert not listed separately |
0x80090304 |
SEC_E_INTERNAL_ERROR: the Schannel code for the TLS internal_error alert |
0x8009030D |
Nothing reachable. A status from the same security package |
0x80090330 |
SEC_E_DECRYPT_FAILURE: the Schannel code for the decryption_failed and decrypt_error alerts |
0x80090329 |
Nothing reachable. A status from the same security package |
Because 0x80090326 is the default mapping, it carries less information than the others. A more specific code, such as an untrusted root or an expired certificate, would tell you where to go. This one tells you the handshake was refused and leaves the reason in the alert, which is why the diagnosis is comparative: what does each machine support, and what has changed since it worked.
The CredSSP settings, as Microsoft publishes them
| Item | Value |
|---|---|
| Policy | Encryption Oracle Remediation |
| Policy path | Computer Configuration, Administrative Templates, System, Credentials Delegation |
| Registry key | HKLM\Software\Microsoft\Windows\CurrentVersion\Policies\System\CredSSP\Parameters |
| Value name | AllowEncryptionOracle (DWORD) |
| Force Updated Clients | 0 |
| Mitigated | 1 |
| Vulnerable | 2 |
| Documented client symptom | An authentication error has occurred. The function requested is not supported |
Keep the last row in view. If the machine in front of you is reporting that the function requested is not supported, you are in the CredSSP story and this table is your fix – patch the machine that is behind, and put the policy back to Force Updated Clients afterwards. If it is reporting 0x80090326, you are in the TLS story and the table is background.
Setting Encryption Oracle Remediation to Vulnerable restores the behaviour the update was written to stop, on every connection that machine makes, not just the one you are trying to fix. If you use it at all, use it briefly and put it back the same day.
Comparing two machines properly
- Run
Get-TlsCipherSuiteon both and compare the lists rather than reading either one in isolation. - Note the Windows build and the last cumulative update on both.
- Check whether either has a security baseline or hardening template applied, and what it set.
- Check the Schannel event entries in the System log on both machines at the moment of the failure.
- Test the same client against a third host, and the same host from a third client, to establish which machine is the odd one out.
Why the error appears on the client when the host is at fault
The client is the side that refuses. It receives the alert, converts it to a code and stops, which is why the message surfaces there whatever the underlying reason. That is also why testing from a second client tells you so much: if two different clients fail the same way against one host, the host is where the work is, regardless of which screen showed the error.
When both machines are current and it still fails
- Check the clock on both. A large skew breaks certificate validation inside the handshake and produces an unhelpful refusal.
- Check whether the host has more than one certificate that could serve the listener and is presenting one you did not intend.
- Check for an endpoint protection product performing local TLS inspection on either machine.
- Check whether the connection is going through an RD Gateway that terminates TLS itself, in which case the gateway is the machine to compare against.
- Read the System log on both for Schannel entries at the time of the failure. They frequently name the alert, which turns this generic code into a specific one.
When a licence is the actual fix
Almost everything here costs nothing: comparing cipher suites, replacing a listener certificate, excluding Remote Desktop traffic from an inspection device, or installing the current cumulative update on whichever machine is behind. The case where a licence really is the answer is a host whose version of Windows Server has left support, because its security stack will never receive another protocol or cipher update and the only way to reach it is to weaken the machine that is still protected. That is spending your remaining protection on one connection rather than fixing anything. Arco supplies Windows Server 2022 Standard along with the client access licences your user or device count needs, and can check what your existing agreement already covers before you buy anything new. If the machine is still in support, patch it and buy nothing.
Every code this article covers
| Code | What it points at | Source |
|---|---|---|
0x80090326 |
SEC_E_ILLEGAL_MESSAGE: the Schannel error code returned for a TLS alert, and the default for any alert not mapped separately. It is not the CredSSP encryption oracle code, which reports that the function requested is not supported | Microsoft Learn |
0x80090304 |
SEC_E_INTERNAL_ERROR: the Schannel code for the TLS internal_error alert | Microsoft Learn |
0x8009030D |
A status from the same security package, seen alongside failed authentication exchanges. Microsoft publishes no meaning for it in a reachable table | not published by the vendor |
0x80090330 |
SEC_E_DECRYPT_FAILURE: the Schannel code for the decryption_failed and decrypt_error TLS alerts | Microsoft Learn |
0x80090329 |
A status from the same security package. Microsoft publishes no meaning for it in a reachable table | not published by the vendor |
Confirm the fix worked
- Connect from the client to the host and confirm you reach the credential prompt and then the desktop.
- Confirm the two machines now share at least one protocol version and cipher suite.
- Check the System log on both machines for Schannel entries during the test and confirm none were written.
- Confirm Encryption Oracle Remediation is back at Force Updated Clients on any machine you loosened.
- Test a second host from the same client, to be sure you fixed the pair rather than weakened the client.
Questions people ask about this
Is 0x80090326 the CredSSP encryption oracle error?
No, and this is the most common misreading of it. Microsoft documents the CredSSP oracle remediation failure as an authentication error saying the function requested is not supported. 0x80090326 is SEC_E_ILLEGAL_MESSAGE, the Schannel code for a rejected TLS alert and the default for any alert not mapped separately.
Is there a free fix?
Nearly always. Comparing cipher suites, replacing the listener certificate, excluding the traffic from a TLS inspection device, or installing the current cumulative update all cost nothing. A licence enters the conversation only when a host is past end of support and will never receive another security update.
Why does the error appear on the client if the host is the problem?
Because the client is the side that refuses. It receives the alert, turns it into a code and stops. Testing from a second client is the quickest way to establish which machine is actually the odd one out.
Should I set Encryption Oracle Remediation to Vulnerable?
Not for this code, and not as a standing configuration for any code. It reopens the exact weakness the CredSSP update was issued to close, on every connection that machine makes. If you use it at all, use it for a short documented window on a machine being fixed the same day.
Can I disable Network Level Authentication instead?
You can, and it will sometimes let the connection through, but it moves the credential prompt inside the session and removes a defence that exists for good reasons. It is a worse workaround than the policy change, not a better one.
