Delinea Secret Server auth bypass: why the FIDO2 hotfix isn't enough
Four critical flaws hit on-prem Delinea Secret Server, two of them letting an outsider sign in as another user. The FIDO2 hotfixes fix one. Here's who needs 12.2.7, how to upgrade, and what to check.
part 9 of 25 · Platform releases, explained from scratchthe ideaGet to 12.2.7, check the keys on file, then fence the login.
after reading, you can- Work out whether your Secret Server is still exposed, even after someone on the team applied the hotfix
- Plan the upgrade to 12.2.7 or later, including step upgrades and the changes that break scripts
- Find FIDO2 keys nobody can explain, and reset them properly instead of just switching FIDO2 off
Delinea has disclosed four critical flaws in on-premises Secret Server, two of which let someone sign in as another user. If you run Secret Server yourself, get to 12.2.7 or later, because that's the only build that closes all four. The quick hotfixes Delinea offered for the FIDO2definitionThe standard behind hardware security keys like a YubiKey. In Secret Server you type your password, then touch the key, and that touch is your second factor. bug leave the other three open, and that's an easy gap to miss.
The vault that holds your admin passwords has a lock that can be picked. Upgrade it this week.
Secret Server is where many IT teams keep the passwords that matter most: domain admins, database owners, service accounts, cloud keys. People sign in to it and it hands them the password they need, or opens the session for them. Delinea found that, under conditions it hasn't fully described, someone could tie their own security key to another person's account and then log in as that person. A second flaw lets a sign-in response from your identity provider be reused to pose as someone else. A third lets an outsider use the server to decrypt or encrypt data with one of its own keys, and a fourth is a booby-trapped link that runs code in a user's browser. Delinea's cloud service isn't listed as affected. If your company runs Secret Server on its own servers, the fix is an upgrade, and the version you pick matters.
Think of a bank's safe-deposit room. You show ID at the front desk, then press your own card to a reader to open the vault. The flaw let a stranger register their card against your box at the desk, so the reader opened it for them. Fixing the desk stops new cards being added. It doesn't remove a card someone already registered, so you also check whose cards are on file.
- Teams running Secret Server on-prem
- PAM and identity admins
- Security and incident response
- IT leads who own the vault
- Owners whose MSP runs it
Why it matters to you
A PAM vault is the most valuable box on most networks. Whoever can sign in as an admin there can usually read the domain admin password, and from there reach almost anything. So a bug that lets an outsider become another user on Secret Server is about as bad as identity bugs get, even before anyone has used it.
Nobody has reported attacks so far. CISA's enrichment of each CVE record marks exploitation as none, none of the four is on CISA's Known Exploited Vulnerabilities list, and Delinea's advisory doesn't mention attacks. That's the good news. The bad news is that CISA also marks the FIDO2 flaw as automatable, which means an attacker could script it against many servers once someone works out the details.
The scores need one caution. Delinea scored the FIDO2 bug with CVSS v3.1 and the other three with CVSS v4.0, so a 9.8 and a 9.5 aren't directly comparable. The v4 vectors for the SAMLdefinitionWhat sits behind most single sign-on. Entra ID or Okta signs a message saying who you are, and Secret Server lets you in because it trusts that signature. and padding oracledefinitionA bug where a server leaks, through its error messages or timing, whether encrypted data it received was well formed. An attacker uses that to decrypt or forge data without the key. bugs say the attack is harder to pull off and needs certain conditions to be present. The XSS bug needs a user to click a crafted link. None of that makes them safe to leave. It tells you which one to worry about first if you can only do one thing tonight: the FIDO2 one.
-
01 The attacker reaches the login page
The attacker only has to get to your Secret Server login page. Going by the CVSS score, they don't need an account, and nobody on your side has to click anything.
-
02 Key registration trusts the wrong person
Under certain conditions, the step that enrols a FIDO2 security key accepts a key for an account the attacker doesn't own.
-
03 Their key now belongs to your admin
From then on, Secret Server accepts the attacker's key as your admin's second factor. They never needed to get hold of the admin's real key.
-
04 They sign in as that user
Delinea says the attacker can then authenticate as the target, with whatever that person can see.
-
05 Secrets and sessions open up
If the target is a Secret Server admin, that means stored passwords, launchers and the settings that control everything else.
For engineers, the 12.2.7 release notes give a few more clues without naming the CVE. They list fixes to the FIDO2 two-factor registration callback, including tighter validation, a corrected challenge comparison and a longer life for the one-time registration key. A wrong challenge comparison is the kind of bug that lets a registration ceremony meant for one session complete for another. Treat that as our reading of the notes, since Delinea hasn't confirmed the root cause.
In Secret Server, FIDO2 is a second factor that comes after the password. That's why this bug is ugly. The key is supposed to be the thing that saves you when a password leaks, and here the key itself was the weak spot. It also matters for cleanup, which we come back to in the playbook: a rogue key stays on the account until someone resets it.
What Delinea fixed, and the hotfix trap
Delinea fixed the FIDO2 bug first, and quietly. Secret Server 12.1.000003 went out with a line saying it fixed CVE-2026-19117, and that details were held back so customers could update before they went public. The CVE record followed about three weeks later, and it lists hotfixes for older lines too. The other three CVEs were published on September 15 and are only fixed in 12.2.7 and later (the XSS bug is also fixed from 12.0.20).
The four Secret Server CVEs at a glance
An attacker registers their own FIDO2 key against a target account and signs in as that user. CVSS v3.1 9.8. Fixed in 12.2.7+, or hotfixes 12.1.3, 12.0.23, 11.9.48, 11.8.2 and 11.7.62.
A valid SAML response from your identity provider can be used to pose as a different Secret Server user. CVSS v4.0 9.5. Affects 10.5.0 to 12.1.3, fixed in 12.2.7+.
Someone with network access and no login can make the server decrypt or encrypt data with one of its keys. The key itself isn't exposed. CVSS v4.0 9.1. Affects 10.5.1 to 12.1.3, fixed in 12.2.7+.
A crafted link runs the attacker's JavaScript in the browser of whoever opens it. CVSS v4.0 9.3, needs a click. Affects 10.2.19 to 11.9.48, fixed from 12.0.20.
There's no FIDO2 hotfix for them. Delinea says to upgrade to a supported version, and anything before 11.9.000025 has to step through 11.9.000006 first.
Delinea lists all four as on-prem issues. The FIDO2 record says it affects on-premises deployments only, so cloud tenants have nothing to install.
Here's the trap. If your team saw the FIDO2 advisory, took the hotfixdefinitionA small update on an older version line that fixes one specific problem without moving you to the next release. for its line and closed the ticket, the job isn't done. The 12.1.3 hotfix fixes FIDO2, but the SAML and padding oracle bugs affect every version from 10.5 up to and including 12.1.3. The 11.9.48 hotfix leaves those two open and the XSS bug as well, since that one runs through 11.9.48. If your server sits on any of those hotfix builds, at least two of the four are still open, and three on 11.9.48 or older.
| Your version | FIDO2 (19117) | SAML (15640) | Padding oracle (15638) | XSS (15639) |
|---|---|---|---|---|
| 12.2.7 or later (incl. 12.3.x) | Fixed | Fixed | Fixed | Fixed |
| 12.1.3 hotfix | Fixed | Open | Open | Not affected |
| 12.1.0 to 12.1.2 | Open | Open | Open | Not affected |
| 12.0.23 hotfix | Fixed | Open | Open | Fixed (12.0.20+) |
| 11.9.48 hotfix | Fixed | Open | Open | Open |
| 11.8.2 or 11.7.62 hotfix | Fixed | Open | Open | Open |
| 10.6.0 to 11.6.x | Open, no hotfix | Open | Open | Open |
| Secret Server Cloud | Not affected | Not listed | Not listed | Not listed |
One cell in that table deserves a footnote you won't find in the advisory. The XSS fix is listed as 12.0.20 or later, but Delinea withdrew 12.0.000020 and 12.0.000021 from distribution, and told customers with an HSM whose server failed after that upgrade to roll back to 11.9.000047. If you're on a 12.0 build, check which one, and plan the move to 12.2.7 or later instead of stopping on 12.0.
How it unfolded
- Aug 11, 202612.1.000003 ships the FIDO2 fix
Delinea kept the details back so customers could update first. If you only patch when an advisory appears, you'd have missed it.
- Aug 28, 202612.2.000007 released
The first build that fixes all four.
- Sep 2, 2026CVE-2026-19117 published
The FIDO2 bypass goes public, with hotfixes for older lines. If you track advisories, this is the date you'd have seen.
- Sep 15, 2026Three more CVEs published
These three only go away on 12.2.7 or later. The FIDO2 hotfixes don't touch them.
- Oct 21, 202612.3.000001 released
- Oct 23, 202612.3.000002 maintenance fix
Fixes a sign-in loop for Windows-auth users with 2FA, introduced in 12.3.000001.
- Oct 25, 2026Still no known exploitation
None of the four is on CISA's KEV list yet. That can change quickly once someone publishes details.
What to do now
What to do now, by who you are
- Ask IT or your managed service provider: do we run Delinea Secret Server, and is it Delinea's cloud or our own servers?
- If it's your own servers, ask for the exact version number in writing. "We patched in September" isn't an answer, because the hotfixes only fixed one of four bugs.
- Ask for a date for the move to 12.2.7 or later, and who signs off the change.
- Read the version off every instance, including disaster-recovery replicas and test servers that hold real secrets.
- Back up the database and the application folder, including encryption.config, then upgrade to 12.2.7 or later. Step through 11.9.000006 first if you're below 11.9.000025.
- Bring the distributed engines and Protocol Handler up to the versions the release asks for. If you skip them, launchers and rotation are what break first.
- Test API scripts and SDK clients after the upgrade, since the direct API authentication default changed.
- Pull every FIDO2 key registration from the audit logs since spring and match each one to a user who actually did it.
- Look for sign-ins to admin accounts from addresses or times that don't fit, and for SAML sign-ins to privileged accounts that never use SSO.
- Search web logs for bursts of failed requests from one source, the noise a padding oracle attack makes.
- Try the login page from a phone off the office Wi-Fi. If it loads, find out who actually needs it reachable from outside.
- Pull a list of who can see the domain admin and database owner secrets. Cut anyone who doesn't need them, and write down why the rest do.
- Put Secret Server's web login behind VPN or zero-trust access so only staff networks can reach it.
- Agree an emergency patch path for the vault, separate from the monthly window.
If you only get one check done today, make it the FIDO2 registrations. Each Secret Server user can have one FIDO2 device registered at a time, so a list of registrations is short and easy to read. Any key registered to an admin account that the admin can't account for is a reason to treat the server as compromised.
# example output. Field and event names vary by version and SIEM, so match them to yours # 1. Every FIDO2 key someone registered, newest first. Each one should match a person you can ask source = secret-server-audit and action contains "FIDO2" | sort -time | table time user ip # 2. Admin sign-ins from addresses that aren't on the office or VPN list action = "LOGIN" and user in ss_admins and ip not in corp_ranges # 3. One address throwing hundreds of failed requests in ten minutes. That's what padding oracle probing looks like source = iis and host = secret-server and status >= 400 | count by src_ip, 10m | where count > 500
The upgrade order that works on-prem
- 01Find every instance
Production, DR replicas, test and the old pilot server someone never turned off. Read the version from each one, not from the change ticket.
- 02Back up the right things
Back up the Secret Server database and the application folder on every web node, with unencrypted copies of encryption.config and database.config. Delinea's checklist says not to start without them, and to plan about two hours.
- 03Step if you have to
Below 11.9.000025, upgrade to 11.9.000006 first. Delinea's notes warn that offline installs which skip this can fail with an integrity check error about the security catalog thumbprint.
- 04Upgrade to 12.2.7 or later
Use the in-product upgrade where you can. If you choose 12.3, go straight to 12.3.000002.
- 05Bring the satellites up
Each release names the engine and Protocol Handler versions it wants. Upgrade them in the same window, or someone finds out on Monday that launchers and rotation have stopped.
- 06Test what breaks
Sign in with each method you use (local, SAML, Windows auth, FIDO2), launch a recorded session, run a rotation and run your API scripts.
- 07Then hunt
With the hole closed, review FIDO2 registrations, admin sign-ins and secret views for the months the server was exposed.
Harden so the next one hurts less
Secret Server has had seven on-prem advisories in about fifteen months on Delinea's own page, and four of them were critical. Plan for more. The goal is that the next login bug costs you an upgrade night and nothing else, because nobody outside your staff networks can reach the login page and a stolen login still can't see much.
Before: the vault's login page answers anyone
Anyone who can load the login page can try their luck, and the only record of it sits on the same server they're attacking.
After: patched, fenced and watched
Only staff networks reach the login, sign-in runs through your identity provider, and every registration and login lands in the SIEM.
- Login page open to the internet, because remote staff once asked for it and nobody took it back
- Patched on the monthly cycle, or when someone notices
- Dozens of people hold Secret Server admin roles, mostly because nobody ever removed them
- Audit logs read only when an auditor asks
- Hotfix taken for the loudest CVE, then the ticket gets closed and nobody looks again
- Login reachable only through VPN or zero-trust access
- An emergency upgrade path for the vault that the team actually rehearses twice a year
- A handful of named admins, with break-glass accounts sealed and watched
- FIDO2 registrations, admin sign-ins and SAML changes alerting in the SIEM
- Someone checks the running version against every open advisory before the ticket is allowed to close
Two of these do most of the work. Getting the login page off the internet turns three of the four CVEs from a remote problem into an insider problem, and the fourth needs someone on your staff to click a link. Alerting on FIDO2 registrations costs almost nothing, because real registrations are rare and each one should match a ticket or a new hire.
One newer fix is worth using while you're in there. Since 12.3.000001, SAML certificate additions and replacements show up in the SAML audit log. Swapping the identity provider certificate is a classic way to make forged sign-ins look valid, so forward that log too.
If you've been breached
If you find a key or sign-in nobody can explain
Work through these in order. The two mistakes to avoid are disabling FIDO2 instead of resetting it, and rotating passwords before the attacker's access is gone.
- 01Containfirst hour
- Cut the login page off from everything except a small admin network at the firewall.
- Back up the database, the application folder with encryption.config, and the IIS and audit logs before you change anything.
- Disable the affected user accounts in Secret Server. Leave them in place so the audit trail stays.
- Open a priority case with Delinea support that cites the CVE numbers.
- 02Investigatefirst days
- Go through the secret audit for each affected account. Every secret it viewed, checked out or launched goes on a list, with the time.
- Trace launched RDP and SSH sessions to the servers they reached, and check those for new accounts or tasks.
- Look for admin-level changes nobody asked for. New users and changed roles are the obvious ones, but check SAML and API settings and folder permissions too.
- Write down which credentials the accounts could reach. That list is your worst case.
- 03Removeday 1 to 2
- Use Reset Two-Factor Authentication on the affected users. Delinea's docs say disabling FIDO2 keeps the device registration, so a rogue key would come back the moment FIDO2 is re-enabled.
- Upgrade to 12.2.7 or later before letting anyone back in, and remove any accounts, API settings or certificates the attacker added.
- 04Rotateafter the upgrade
- Rotate every password and key the affected accounts could read, starting with accounts that can reset other accounts.
- Use Secret Server's own rotation where it works and confirm heartbeat afterwards. Do the rest by hand from a list.
- Revoke and reissue API tokens and SDK client registrations that the compromised accounts could see. Scripts will break for a day, which is the point.
- 05Report and learnfirst week
- Bring in legal early. Depending on your sector and what the credentials opened, regulators or customers might be owed a notice, and the clock often starts at discovery.
- Call your cyber insurer early. Most policies have a notice window, and a late call can make the claim harder.
- Fix what let it happen: the open login page, slow upgrades, too many admins, no alerting on registrations.
A word on the padding oracle bug, because it changes what you look for. Delinea says the server's key isn't exposed, but an attacker could decrypt or encrypt data with it. Microsoft's guidance on this class of bug describes the attacker decrypting a message byte by byte by sending slightly changed versions and watching how the server responds. That takes a lot of requests. A burst of thousands of failed requests from one address, sitting in IIS logs that nobody watches, is the thing to search for.
What this could look like: three typical setups
What this could look like: three typical setups
- the setup
- A logistics firm moved to Delinea's cloud service two years ago. Admins sign in through Entra ID with MFA.
- what happens
- Delinea lists these as on-prem issues, so there's nothing for the firm to install. Their old on-prem server was switched off after the move, or so the team believes.
- what to do
- Confirm the old server really is off and its database is gone or locked away. Review admin accounts and FIDO2 registrations anyway, since that's cheap.
- the setup
- A regional hospital runs Secret Server 12.1 on-prem. In September the team applied the 12.1.3 hotfix and closed the ticket. Staff sign in through SAML.
- what happens
- FIDO2 is fixed, but 12.1.3 is still inside the range for the SAML impersonation and padding oracle bugs. The hospital uses SAML, so the bug in the path it relies on every day is still open.
- what to do
- Plan the move to 12.2.7 or later this week, test SAML and launchers after it, then review SAML sign-ins to privileged accounts since the summer.
- the setup
- A manufacturer still runs Secret Server 11.6 on a server that only answers from the plant's admin network. Nobody has upgraded it in years.
- what happens
- There's no hotfix below 11.7, so every one of the four bugs is open. The firewall keeps outsiders away, but anyone already on the admin network, or anyone who clicks a crafted link, can still use them.
- what to do
- Schedule the step upgrade through 11.9.000006 to 12.2.7 or later, with a database and encryption.config backup first. Keep the network fence, it's doing its job.
These three are made up. None of them is a real incident or a Cohash client. They're here to show how the same advisory plays out differently depending on where Secret Server runs and which build it's on.
How Cohash helps
We implement and upgrade Delinea Secret Server, including the step upgrades, distributed engines and launchers that make these jobs slow. If you want a second pair of hands for the move to 12.2.7 or later, or a review of who can see what in your vault afterwards, start with our Delinea page, look at Delinea engineers you can bring in, or take the free assessment to see where your privileged access stands.
What to take from this
- 01
Self-hosted Secret Server needs 12.2.7 or later. That's the only build that fixes all four September CVEs.
- 02
The FIDO2 hotfixes (12.1.3, 12.0.23, 11.9.48, 11.8.2, 11.7.62) leave the SAML and padding oracle bugs open.
- 03
Secret Server Cloud isn't listed as affected. Check that no old on-prem server is still running.
- 04
No exploitation is known yet, but CISA rates the FIDO2 bug as automatable. Don't wait for the monthly window.
- 05
Review FIDO2 registrations on admin accounts. Reset two-factor on anything you can't explain, don't just disable it.
- 06
Take the login page off the internet, so the next login bug is an upgrade night and not an incident.
01Am I affected by CVE-2026-19117?
If you run Secret Server on your own servers at a version from 10.6.0 up to 12.1.2 and haven't applied one of Delinea's hotfixes, yes. The hotfixed builds are 12.1.3, 12.0.23, 11.9.48, 11.8.2 and 11.7.62, and 12.2.7 or later fixes it too. Delinea's cloud service isn't affected.
02We applied the FIDO2 hotfix when it came out. Are we done?
Probably not. The hotfixes fix only the FIDO2 bug. The SAML impersonation and padding oracle bugs affect versions up to 12.1.3, so every hotfix build is still exposed to them. Upgrade to 12.2.7 or later.
03We're on Secret Server Cloud. Do we have to do anything?
Delinea lists all four as on-prem issues, so there's no upgrade to run. It's still worth checking that any old on-prem server from before the move is switched off and that admin accounts and FIDO2 registrations look right.
04Has anyone exploited these?
Nothing public so far. CISA's records for all four CVEs mark exploitation as none, they aren't in the Known Exploited Vulnerabilities catalog, and Delinea's advisory doesn't mention attacks. CISA does mark the FIDO2 bug as automatable, which is why it shouldn't wait.
05Should we upgrade to 12.2.7 or straight to 12.3?
Either fixes all four. If you pick 12.3, use 12.3.000002, because 12.3.000001 sent Windows-auth users with two-factor back to the login page. Check the distributed engine and Protocol Handler versions each release needs before you start.
06We don't use FIDO2 keys. Does the FIDO2 bug still matter?
Delinea hasn't said whether the bug needs FIDO2 to be switched on for the target account, so don't assume you're safe. Upgrade regardless, since the other three bugs don't depend on FIDO2 at all.
07Our version is older than 11.7. What now?
There's no hotfix for it, so you upgrade. Anything before 11.9.000025 goes to 11.9.000006 first, then on to 12.2.7 or later. Back up the database and the application folder, including encryption.config, before each step.
Three questions, about a minute
Pick an answer and we'll tell you if it's right, and why it is or isn't. The score stays in this browser.
-
Q1Your on-prem Secret Server runs the 12.1.3 hotfix you applied in September, and staff sign in through SAML. What's your status?
show the answer
B. Fixed for FIDO2, still exposed to the SAML and padding oracle bugs The 12.1.3 hotfix fixes CVE-2026-19117 only. The SAML impersonation and padding oracle bugs affect versions up to and including 12.1.3, so the fix is 12.2.7 or later.
-
Q2An admin swears they never registered the FIDO2 key that's on their Secret Server account. What do you do about the key?
show the answer
D. Use Reset Two-Factor Authentication, after containing access and saving logs Delinea's docs say disabling FIDO2 keeps the device registration, so the rogue key would return when FIDO2 is re-enabled. Resetting removes it. Contain and preserve evidence first, and rotate once the attacker's access is gone.
-
Q3You moved to Secret Server Cloud last year and the old on-prem server was supposedly switched off. What's worth checking?
show the answer
B. Confirm the old on-prem server and its database are really gone or locked away Delinea lists these as on-prem issues, so the cloud tenant needs no upgrade. A forgotten on-prem server, or a copy of its database, would still carry every bug.