Deep Dive of the Day After the vCenter zero-day: patching is step one, then check your SSO domain and STS signing certificate
This week's vCenter flaw is unusual because of what attackers did with it, not just how bad it is. CVE-2026-123890 sits in the VMware Directory Service (vmdir), the LDAP-style store behind vCenter Single Sign-On. According to the reporting on Mandiant's findings, the attackers' tool, BONESAW, used it to get an [email protected] session, copy the directory database, pull out SAML signing material and create a new SSO administrator account, as The Hacker News and this write-up describe. Installing 9.1.1.0200, 9.0.2.0200 or 8.0 U3m closes the hole. It doesn't remove an account someone created or make stolen keys useless, so it's worth understanding the pieces.
Start with the Security Token Service. Broadcom's 9.0 documentation describes it as "a Web service that issues, validates, and renews security tokens". When you log in, STS checks your credentials and builds a SAML token; it signs that token with a private key, and the other vCenter services check the signature against the published public certificates. Those STS signing certificates are stored in vmdir, which is why a flaw in vmdir matters so much.
That's also why a stolen signing key is serious. With it, anyone can mint tokens the services will trust, which is the "Golden SAML" technique the second tool, SAWSNAKE, used to get administrator session cookies, per the write-up. A forged token doesn't need a password and doesn't go through your identity provider's MFA, so it bypasses the controls you'd normally rely on.
The fix for a stolen key is a new key. The 9.1 procedure is short: log in as [email protected] (or another member of the SSO Administrators group), go to Administration, then Certificates, then Certificate Management, select the certificate on the STS Signing tab and choose "Refresh with vCenter Certificate". VMCA issues a new certificate, replaces the old one in vmdir, and STS signs new tokens with it. The page says the new certificate is pushed to linked vCenters too, and that you don't need to restart this vCenter or the others in Enhanced Linked Mode.
There are traps. If you use a custom or third-party STS certificate for compliance, the refresh overwrites it with a VMCA one, and the docs warn that may put you out of compliance; the alternative is "Import and Replace" with a new certificate of your own. If the client offers "Force Refresh" instead, stop and read: the page says it requires restarting every vCenter and "can render those systems inoperable until you do so". KB 318197 adds one sensible step before any of this: take offline (cold) snapshots of all the vCenters in the SSO domain.
Next, look at who's in the SSO domain. Check the members of the Administrators group and any local SSO users you don't recognise, and compare them with what you expect; the attackers' account in Mandiant's case was an SSO administrator. The reported chain also used the VAMI API to create a local appliance administrator and switch SSH on, and added an SSH key to ESXi root, so check VAMI's SSH setting, the local accounts on the appliance and the authorised keys on your hosts.
Be careful with the timeline. Mandiant believes the group knew about this flaw before VMSA-2026-0006 was published in July, and SecurityWeek reports BONESAW was used against a vCenter already patched for CVE-2026-59309, per SecurityWeek. Broadcom's FAQ confirms the two flaws are distinct. So "we patched in August" doesn't mean "we were safe in September"; if a vCenter was reachable from a network an attacker could have been on, the checks above are worth doing whatever your patch history.
For production, sequence it: patch first (it isn't Quick Patch-eligible, so plan for the management outage; running workloads carry on, per Rapid7), then review SSO administrators, VAMI and host SSH settings, then refresh the STS signing certificate if there's any doubt. Forward vCenter and ESX logs to somewhere an attacker can't edit them; both write-ups note the group cleaned up after itself, and Brandefense relays BeyondTrust's point that local logs were deleted on the appliance they came in through.
If you're running VCF, remember that SDDC Manager and VCF Operations hold service accounts and certificates for vCenter as well. Refreshing the STS certificate is a vCenter operation and the docs don't call out a VCF-specific step, so in a production fleet check the VCF certificate and credential views afterwards and try it in a lab first.
In a nested lab, this is a cheap rehearsal: patch a lab vCenter to 9.1.1.0200, add a throwaway SSO administrator so you know where to find it, then refresh the STS certificate and watch what happens to open vSphere Client sessions, PowerCLI connections and anything else that authenticates. It's much better to learn that before you need to do it on a vCenter you're worried about.
Based on Refresh a vCenter STS Certificate Using the vSphere Client (vSphere 9.1) Official Managing the vCenter Security Token Service (vSphere 9.0) Official KB 318197: STS signing certificates about to expire Official VMSA-2026-0009 questions and answers Official ARMORY CASTLE's VMware vCenter attack chain (NetmanageIT)