decrypted · 21 november 2026 · vulnerabilities and patching · ai and llm security · digital sovereignty

A VMware vCenter flaw handed attackers the master key

Broadcom disclosed a critical flaw in VMware vCenter on Thursday, and Google's Mandiant says a suspected Chinese state-linked group had been exploiting it since early September. vCenter is the console administrators use to run whole estates of virtual servers, so this is a story about who holds the keys to the building. The flaw, CVE-2026-123890, scores 9.8 out of 10 on CVSS, the severity scale, and Broadcom lists no workaround. This is new since this morning's post.

How the break-in worked

Mandiant says the bug sits in vCenter's directory service. An attacker with network access to the appliance, and no password, can bypass the login and arrive as the built-in administrator. Think of a hotel where the front desk hands over the master key to anyone who asks in just the right way. No forced door, no alarm: the manager has simply walked in.

The attackers did more than visit. Mandiant describes a tool, Bonesaw, that copies the directory database and extracts the key used to sign single sign-on tokens. Single sign-on, or SSO, is one login for many systems, and SAML is the standard behind many such logins. With that signing key, the group can forge passes for any user and create an administrator account of its own. Patching afterwards closes the door but does not take back a copy of the key.

What defenders can see

Look for unexpected administrator accounts whose names begin "sps-", and for SSH, the encrypted remote login, being switched on through the appliance's management interface and then off again. Attackers scanned for ports 2012, 5480 and 9080. Mandiant also publishes file names, network addresses and detection rules. Its post does not say whether to rotate signing material after a suspected compromise. Given that the key was the prize, that is a question we would put to Broadcom before assuming a patched server is a clean one.

The fixed versions are 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. For vCenter 7.0, Broadcom's advisory says only to contact support if you have an extended support contract. SecurityWeek notes that 7.0 left general support in October 2025.

The Secure by Design lesson

First, a management console should not be reachable from places that do not need it. Mandiant's own advice is to restrict ports 2012, 22, 5480 and 443 to allowlisted addresses and to block outbound traffic from the appliance. That is an afternoon's network work. SecurityWeek says the group got in through a separate remote support flaw: the control panel was one hop from the outside world.

Second, one secret should not be able to impersonate everyone. A signing key that, once copied, yields permanent administrator passes is a design choice. Buyers can ask vendors how that key is protected and how quickly it can be replaced.

Third, end of support is a security event: a UK organisation on 7.0 without an extended contract must pay for cover or migrate, and directors should hear that as a risk, not an IT upgrade.

Also this week

An AI tool in the European Commission's attack. The Commission confirmed on Friday that an AI-assisted cyberattack, detected on 22 October and confirmed on 26 October, disrupted several of its applications, most of which have since been restored. Spokesperson Thomas Regnier said there was no evidence of a data breach affecting staff, and named neither an attacker nor the tool. Earlier this month, CERT-EU warned that threat actors worldwide were using ARTEX, an open-source framework that turns a plain-English objective into scanning and exploitation, and that it has "no hard technical enforcement" on scope. SecurityWeek links the two only implicitly, and so do we. The lesson is speed. CERT-EU's advice is emergency patching of internet-facing systems, isolating management panels from the internet, and logging that covers AI tools too, which is the vCenter advice again. Opinion, not advice: expect the gap between disclosure and exploitation to keep shrinking, so a patching process that takes a fortnight is now a design flaw.

Nvidia and an unproven ransomware claim. The Gentlemen ransomware group has listed Nvidia on its leak site. The listing carries a logo and a generic description but no proof, and SecurityWeek says it does not establish that Nvidia was breached. Nvidia said it is investigating "an incident that may have affected a third-party partner" and has "found no indication that Nvidia systems or networks were affected." The group's leak site lists more than 1,100 alleged victims, which tells you how busy it is, not whether this entry is true. Treat this as a claim, not a fact. The lesson for UK boards is that when a supplier is compromised, your name can appear on someone else's leak site, so know which suppliers hold your data, what they must tell you, and how fast. Secure by Design here means contracts and architecture that limit what any one partner can hold.

Sources

If you run vCenter or depend on suppliers who do, get in touch and we will help you work out where you stand.

More like this

Get the next post by email: subscribe to Decrypted. Double opt-in, unsubscribe any time, or take the RSS feed.

Prefer to listen? Decrypted on Apple Podcasts, or paste the podcast feed into any app.

Play here Loads Apple's player when you press it.