Disabled is not always disabled: what the Nexus Repository SSO bug teaches us about offboarding
- What happened?
- Why this is an IAM lesson, not just a patch note
- Are you affected?
- What you should do now
- Conclusion
Dear readers,
it’s Sunday, so let’s start with a small quiz. An administrator opens an application, finds a user and clicks “disable”. Can that user still sign in?
If your answer is “of course not, that’s what the button is for”, then today’s post is for you. In CVE-2026-113373, the answer for Sonatype Nexus Repository 3 Pro was “yes, as long as they come in through single sign-on”. It’s not an identity provider bug this time, but it is a very clean example of something that goes wrong in many applications behind an identity provider.
What happened?
On November 4, 2026, Sonatype published the advisory CVE-2026-113373: Disabled or Locked Single Sign-On Users Can Sign In Again. The summary is refreshingly short: Nexus Repository 3 Pro “does not check the status of a local user account” when the user signs in through a configured SAML or OAuth2/OpenID Connect identity provider.
So the sequence looks like this:
- An administrator disables or locks an account in Nexus Repository.
- The person behind it still has valid credentials at the identity provider.
- They sign in via SSO again and receive a new session with the roles of that account.
No exploit code, no crafted request, no percent-encoded magic. Just a normal login that should have been refused. According to the advisory, that can mean read access to private repositories, the ability to publish components, or administrator functions, depending on your role mapping. Sonatype rates it CVSS 4.0 9.1 (Critical) and classifies it as CWE-1390 (Weak Authentication). The NVD analysis shown at OpenCVE scores it 8.1 (High) under CVSS 3.1. Sonatype also states that it is not aware of active exploitation at the time of the announcement.
Why this is an IAM lesson, not just a patch note
In a federated setup there are usually two places where an account can be switched off:
- At the identity provider: the user cannot authenticate anywhere anymore.
- In the application: the user can still authenticate, but this one application should say “not you”.
The second switch is used more often than you might think. A contractor moves to another project, an account is locked during an investigation, or the application owner simply has no rights in the IdP and needs a quick fix on a Friday afternoon.
The bug is that the SSO login path treated a valid assertion or token as the whole answer. But the IdP only answers the question “who is this?”. The question “is this account still allowed in here?” belongs to the application, and it has to be asked on every login path: local password, API token, SAML, OIDC. If one path skips the check, the disable button is only a decoration for that path.
(It’s a bit like taking back someone’s office key while their badge still opens the side entrance. Technically you did something. Practically, they are at the coffee machine.)
Are you affected?
The advisory lists four preconditions, and all of them must be true:
- You run Nexus Repository 3 Pro in version 3.86.0 up to and including 3.97.x.
- A SAML or OAuth2/OpenID Connect realm is enabled.
- An account for the user exists in Nexus Repository and an administrator has disabled or locked it.
- The user still has valid credentials for the same account at the identity provider.
The anonymous user is not part of this attack path. So this is not an “anyone on the internet” problem. It’s a “people you thought you had locked out” problem, which is uncomfortable enough.
What you should do now
Sonatype’s recommendations, plus a few thoughts from my side:
- Upgrade to Nexus Repository 3 Pro 3.98.0 or later. Have a look at the 3.98.0 release notes before you do, because they describe changed defaults (for example around script execution and audit events) and already point to a 3.98.1 patch release.
- Review your logs after upgrading. Sonatype advises to look for sign-ins by accounts that an administrator had disabled or locked, and to treat any activity of those accounts as suspect.
- Can’t upgrade yet? Then disable the account in your identity provider, or remove it from the groups that map to Nexus Repository roles. If you cannot change the IdP, the advisory mentions deactivating the SAML or OIDC realm until you upgrade. That stops all SSO logins, so choose wisely.
- Test your other applications. This part is my own advice, not Sonatype’s. Pick an important application behind your SSO, disable a test account locally and try to log in via SSO. Then try an old API token of that account. You might be surprised.
- Make the IdP the primary off-switch. In your offboarding process, a local “disable” in an application should be an additional step, never the only one. If you provision with SCIM, check that a deactivation in the IdP really arrives in the application.
Conclusion
Authentication is not authorization, and neither of them is lifecycle. An identity provider can tell an application who is knocking. Whether the door should open is still the application’s decision, on every door it has. CVE-2026-113373 is fixed with an upgrade, but the question behind it is worth asking for every application in your landscape: when we click “disable”, which login paths actually listen?
Have a relaxed Sunday, and maybe test one disable button tomorrow!
Magikarp
This post was researched and written by an AI agent (Claude) in Magikarp’s style and published automatically.