What is vCenter — the key to every VM, hit by a zero-day
VMware vCenter Server is the central management server of a VMware virtualization environment. It manages the ESXi hosts and every virtual machine running on them from one place, and it also holds the single sign-on (SSO) service and directory that decide who is allowed to create, stop or copy those machines. Administrator rights on one vCenter are therefore administrator rights over everything beneath it. On November 19, 2026 (US time) Broadcom published security advisory VMSA-2026-0009 for CVE-2026-123890, an authentication bypass in vCenter rated 9.8 under CVSS 3.1, with fixes in 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. There is no public fix for 7.0 and no workaround. Google Threat Intelligence Group (GTIG) says a suspected China-nexus espionage group it tracks as ARMORY CASTLE had been using the flaw as a zero-day since early September 2026. The Broadcom advisory does not mention exploitation, and as of November 20, 2026 the flaw was not in the Known Exploited Vulnerabilities (KEV) catalog kept by CISA.
The three lines
- Definition — vCenter is the central server that manages ESXi hosts and virtual machines, and it contains the single sign-on service and the directory service (vmdir). The flaw is in vCenter, not ESXi
- The flaw — CVE-2026-123890, CVSS 9.8, disclosed November 19, 2026. Fixed in 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. No public fix for 7.0, no workaround
- Exploitation — Google GTIG reports zero-day use from early September 2026. The Broadcom advisory is silent and the flaw is not on KEV as of November 20, 2026. Patching does not invalidate keys that have already been stolen
Key questions
- What is VMware vCenter
- **The central management server of a VMware environment. It manages ESXi hosts, virtual machines and the accounts that control them from one place.** | Component | What it does | Relation to vCenter | |---|---|---| | ESXi | Hypervisor installed on each physical server; actually runs the VMs | vCenter manages many of them together | | Virtual machine (VM) | A virtual server running on ESXi | Create, stop and copy commands come from vCenter | | SSO | Single sign-on for administrators | Lives inside vCenter | | vmdir | Directory service holding accounts and permissions | Where CVE-2026-123890 sits |
- CVE-2026-123890 vCenter fixed version
- **vCenter 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. There is no public fix for 7.0 and no workaround.** | Branch | Fixed version | Note | |---|---|---| | vCenter 9.1.x | 9.1.1.0200 | Build 25901347 | | vCenter 9.0.x | 9.0.2.0200 | Build 25901331 | | vCenter 8.0 | 8.0 U3m | Build number given by one source only | | vCenter 7.0 | No public fix | Extended-support customers contact Broadcom | | VCF 5.x | vCenter 8.0 U3m plus SDDC Manager 5.2.5 | Both are needed | | ESX (ESXi) | Not affected | Per Broadcom |
- Is the vCenter zero-day being exploited
- **Google GTIG reports exploitation since early September 2026. The Broadcom advisory does not mention exploitation, and the flaw is not in CISA KEV as of November 20, 2026.** | Party | Position | When | |---|---|---| | Google GTIG / Mandiant | ARMORY CASTLE (formerly UNC5221) used it as a zero-day | From early September 2026 | | Broadcom | Advisory does not mention exploitation; credits Mandiant as reporter | November 19, 2026 | | CISA | Not on KEV; SSVC entry on the CVE record says exploitation none | November 19–20, 2026 | | Public exploit code | None confirmed | November 20, 2026 |
VMware vCenter Server is the central management server of a VMware virtualization environment. It is where administrators create, stop and move hundreds or thousands of virtual machines running on ESXi, the hypervisor installed on each physical server. The single sign-on service those administrators log in to, and the directory of their accounts, live inside the same server. Administrator rights on vCenter are not rights over one machine. They are rights over every virtual server beneath it.
On November 19, 2026 (US time), an authentication bypass in that server was disclosed as CVE-2026-123890. The Broadcom advisory is VMSA-2026-0009 and the score is 9.8 under CVSS 3.1. Google Threat Intelligence Group (GTIG) says a suspected China-nexus group had been using the flaw since early September 2026, more than two months before a fix existed. This explainer covers what vCenter is, why attackers go for it first, which versions fix the current flaw, and what is left to do after patching, all from the defender's side.
1. What vCenter does — ESXi runs the servers, vCenter runs ESXi
Virtualization splits one physical server into many virtual ones. The software that does the splitting is a hypervisor, and in the VMware product line that is ESXi. ESXi is installed once per physical machine: ten servers, ten copies of ESXi.
Past a few dozen hosts, nobody can manage them by logging in to each one. vCenter is the management server that ties those ESXi hosts together. Help Net Security, Rescana and SOC Prime all describe it as the central management plane of a VMware environment, the place that manages hosts and virtual machines and holds the logins, directory and single sign-on that control them.
| Component | Where it lives | What it does | Relation to the current flaw |
|---|---|---|---|
| Virtual machine (VM) | On an ESXi host | The virtual server where business systems actually run | Not the direct target; reachable once vCenter falls |
| ESXi (ESX) | On each physical server | Hypervisor that runs the VMs | Broadcom says it is not affected |
| vCenter Server | One management server (or a few) | Central management of many ESXi hosts and VMs | The affected product |
| SSO (single sign-on) | Inside vCenter | One login for administrators across the environment | An administrator session here is what the attacker obtains |
| vmdir (directory service) | Inside vCenter SSO | Stores accounts and permissions | The affected component |
| SAML signing key (STS) | Inside vCenter | Signs login tokens so other components trust them | If stolen, forged administrator tokens become possible |
| VAMI (appliance management interface) | Inside vCenter | Settings of the vCenter appliance itself, such as turning SSH on or off | Where post-intrusion setting changes leave traces |
In one line: ESXi runs the virtual servers, vCenter drives ESXi, and SSO with vmdir decides who may drive vCenter. The current flaw sits in that last layer, the one that checks who you are.
Our October 12, 2026 explainer, "What hypervisor ransomware is — lock the layer below and hundreds of servers stop at once," dealt with the ESXi layer, where virtual disk files are encrypted wholesale. This piece is about the layer above it: the management console and the identity system. In that article's attack sequence, step two was taking the management plane. vCenter is that management plane.
2. Why one compromised vCenter endangers everything — session, signing key, hosts, disks
With an ordinary server flaw, the damage is the one server that was hit. vCenter is different because its job is to drive other servers. Rescana, ThreatDefence and SOC Prime all put it the same way: control of vCenter is control of the ESXi hosts it manages and the virtual machines on them.
The table below lists what the attacker gained after reaching vCenter in the case GTIG reported, as relayed by our sources and kept at the level a defender needs.
| What the attacker gained | Its legitimate purpose | What follows if it is taken | Source |
|---|---|---|---|
| An administrator session | The logged-in state of the top administrator account ([email protected]) | Full control of vCenter with no password | Rescana, ThreatDefence, SOC Prime |
| Directory data | The master copy of accounts and permissions | Account information is read in bulk | Help Net Security, Rescana, SOC Prime |
| A new SSO administrator account | The normal way to add an administrator | A back door that survives a password change on the original account | Help Net Security, Rescana, SOC Prime |
| The SAML signing key | Signs login tokens to prove they are genuine | Forged administrator tokens, bypassing multi-factor authentication | Help Net Security, Rescana, SOC Prime |
| The vCenter master key and database credentials | Protects secrets vCenter stores | Stored secrets can be decrypted | SOC Prime |
| Access to ESXi hosts | The channel vCenter uses to command hosts | A way to log in to hosts directly | Rescana, SOC Prime |
| Guest disks | The storage of the virtual machines | Disks are read without logging in to the VM | Rescana, ThreatDefence, SOC Prime |
The last row shows what a vCenter compromise really is. According to Rescana and ThreatDefence, the attacker pulled the Windows domain account database (NTDS.dit) and registry hives out of a guest virtual machine's disk. Security software normally runs inside the virtual machine, and reading the disk from the layer outside it is not visible from inside. That is why a breach of the virtualization layer puts the account system of the whole organization at risk, not just the virtualization team's.
The SAML signing key deserves its own note. Single sign-on issues a token that says "this person is an administrator," and other components trust the signature on it. Whoever holds the signing key can mint that token. Help Net Security and Rescana report that one of the attacker's tools can forge SAML tokens to bypass multi-factor authentication and impersonate users. In practice that means access that survives both a password reset and an MFA requirement.
The same structure recurs across management tools. Configuration management servers and remote management tools occupy the same position: one machine that moves hundreds. A system built to move many machines at once gives an attacker the same leverage.
3. CVE-2026-123890 and the fixed versions
What the four outlets relay from the Broadcom advisory and the CVE record:
| Item | Detail | Source |
|---|---|---|
| Advisory | VMSA-2026-0009, published November 19, 2026 (US time) | All four |
| CVE | CVE-2026-123890 | All four |
| Score | 9.8 under CVSS 3.1, Critical | All four |
| Type | Authentication bypass using an alternate path or channel (CWE-288) | Rescana |
| Location | VMware Directory Service (vmdir) in vCenter SSO | Rescana, ThreatDefence, SOC Prime |
| Preconditions | Network access to vCenter. No account, no user interaction | Help Net Security, Rescana, ThreatDefence |
| Port | TCP 2012 per GTIG. Broadcom names no port | Rescana, ThreatDefence, SOC Prime |
| Result | An authenticated session as the top administrator account | Rescana, ThreatDefence, SOC Prime |
| Reporter | Mandiant | Help Net Security, Rescana, ThreatDefence |
| Workaround | None | All four |
A 9.8 is what results when network reach, low complexity, no privileges and no user interaction all line up. How to read such scores is covered in our October 6, 2026 piece, "What CVSS is — why GitLab's critical flaw scored 9.9, not 10."
Accounts of the mechanism differ in detail. Rescana and SOC Prime, relaying GTIG, describe a flaw in the authentication handshake that leaks a session key without a password. On root cause, Rescana points to predictable timestamp generation, ThreatDefence to a predictable UUID v1 value, and SOC Prime to weaknesses in cryptographic logic and authorization. Broadcom itself says only that a party with network access to vCenter may bypass authentication. The conclusion for a defender is the same either way: anyone who can reach vCenter over the network can become its administrator without an account.
The fixed versions:
| Product / branch | Fixed version | Build | Note |
|---|---|---|---|
| vCenter 9.1.x (VCF/VVF) | 9.1.1.0200 | 25901347 | Build number per Rescana |
| vCenter 9.0.x (VCF/VVF) | 9.0.2.0200 | 25901331 | Build number per Rescana |
| vCenter 8.0 | 8.0 U3m (8.0.3.01200) | 25934264 | Build number from Rescana only; ThreatDefence gives none |
| vCenter 7.0 | No public fix | — | Extended-support customers contact Broadcom |
| VCF 5.x | vCenter 8.0 U3m plus SDDC Manager 5.2.5 | — | Both must be applied |
| Telco Cloud Platform / Infrastructure | Per a separate Broadcom knowledge base article | — | Rescana, ThreatDefence, SOC Prime |
| vCenter 6.5, 6.7 | Presumed affected | — | As relayed from the Broadcom FAQ; fix status not confirmed |
| ESX, NSX, VCF Operations, VCF Automation, Workstation, Fusion | Not affected | — | ThreatDefence; Rescana agrees on ESX |
Four things in that table are easy to trip over.
First, there is no workaround. All four outlets say Broadcom offers none. Per Rescana, the Broadcom FAQ says only that compensating controls such as firewalls may help and leaves the decision to each customer.
Second, there is no fast-patch route. Rescana and ThreatDefence report that this fix is not eligible for vCenter Quick Patch. It takes a full update, and the fixes are cumulative. There is also a caution that installing 8.0 U3m or 9.0.2.0200 blocks an upgrade to VCF 9.1.1, so organizations that were planning a move to a newer release should recheck their sequence.
Third, scanners may miss 7.0. Rescana points out that the CVE record and the US National Vulnerability Database list no 7.0 entry, which could lead a vulnerability scanner to mark 7.0 as unaffected. No fix and not affected are different things. Why out-of-support branches stop receiving fixes is the subject of our October 26, 2026 piece, "What is end of life (EOL) software? Why PHP2Shell fixes stopped at 8.2 — and 8.2 ends December 31, 2026."
Fourth, the same advisory fixes four other flaws. According to ThreatDefence these are an SQL injection in SDDC Manager (CVE-2026-123891, 8.1), an authenticated command injection (CVE-2026-123892, 7.2), a local privilege escalation (CVE-2026-123893, 6.7) and a SAML token signature bypass in VMware Tools for Windows (CVE-2026-123894, 6.7), with VMware Tools fixed in 13.1.5.0 and 12.5.5. Those version details come from ThreatDefence alone.
4. Exploitation — Google says since early September; the Broadcom advisory says nothing
The part of this story that needs the most careful reading is that the answer to "was it actually used" depends on whom you ask.
| Party | What it says | When |
|---|---|---|
| Google GTIG / Mandiant | A suspected China-nexus espionage group, ARMORY CASTLE (formerly UNC5221), used the flaw as a zero-day | Exploitation from early September 2026; report dated November 19–20 |
| Broadcom | Advisory does not mention exploitation; credits Mandiant as reporter | November 19, 2026 |
| CISA | Not in the KEV catalog; SSVC entry on the CVE record says exploitation none | Value as of November 19; not on KEV through November 20 |
| Security outlets and vendors | No public exploit code; the attacker's tool has not leaked | As of November 20, 2026 |
As relayed from GTIG, the attacker did not come in through vCenter. Entry was through a zero-day in a remote support product, BeyondTrust Remote Support (CVE-2026-102764), followed by movement to internal Linux appliances that had no endpoint detection, and from there to vCenter. Help Net Security notes the group has used zero-days and known flaws against edge and security appliances since 2023.
One point follows directly. This attack worked even though vCenter was not the internet-facing door. Once there was a foothold somewhere inside, vCenter was the next objective.
The flaw has a prequel. On July 29, 2026 Broadcom fixed CVE-2026-59309, another authentication bypass in the same directory service, under advisory VMSA-2026-0006. That one used LDAP traffic, and Broadcom calls CVE-2026-123890 "a new and distinct issue." According to GTIG, however, the attacker's tool was run against a vCenter that had already been patched for the July flaw. The tool carries a build timestamp of July 13, 2026, a value the outlets note can be faked. Help Net Security, Rescana and SOC Prime present this as evidence that the attacker knew about the flaw before the July patch.
Why a zero-day, a flaw used while no fix exists, is the most expensive tool in security is covered in our August 10, 2026 piece, "Zero-day explained — why an unpatched flaw is the most expensive thing in security." This case also shows the other side. Now that a fix is public, other attackers will take it apart to find the same flaw. That process is explained in our November 18, 2026 piece, "What is patch diffing — why a fix is a map, 63 days to −7."
Absence from KEV should not be read as "not exploited." KEV lists what CISA has confirmed and added, and additions lag. Rescana flags the mismatch between the SSVC value and the GTIG report as unresolved.
5. Why patching is not the end — stolen keys stay valid
A fix closes the door. It does not remove someone who is already inside. Rescana says so explicitly: if the attacker has already taken the SAML signing key and the master key, those keys remain valid after patching, and they have to be rotated.
Go back to the table in section 2 and the reason is visible. Of everything the attacker gained, only the first row, the administrator session, depends on the flaw itself. The rest, the added administrator account, the signing key, the access planted on hosts, keeps working after the flaw is closed. And if exploitation began in early September, the period to review runs for more than two months before the November 19 fix.
The checks below group what Rescana, ThreatDefence and SOC Prime relay from the GTIG report. Detailed indicators such as file hashes and detection rules are in the GTIG report and are not reproduced here.
| Order | What to check | Why | Source |
|---|---|---|---|
| 1 | vCenter version | At or above 9.1.1.0200, 9.0.2.0200 or 8.0 U3m. On 7.0 there is no fix | All four |
| 2 | Who can reach TCP 2012 | The port the attack reportedly used. Narrow it if reachable from outside the management network | Rescana, ThreatDefence, SOC Prime |
| 3 | The list of SSO administrator accounts | Unknown local administrators, especially any created since September | Help Net Security, Rescana, SOC Prime |
| 4 | The history of SSH setting changes on vCenter | The attacker reportedly enabled SSH and later disabled it | Rescana, SOC Prime |
| 5 | Root SSH keys and login records on ESXi hosts | Whether a direct way in to the hosts was planted | Rescana, SOC Prime |
| 6 | Temporary directories and unfamiliar executables on the vCenter appliance | Where tools were reportedly placed | Rescana, ThreatDefence |
| 7 | If there are signs of compromise, rotate keys and credentials | Signing and master keys are not invalidated by the patch | Rescana |
A few additions.
On item 4, SOC Prime and Rescana give the same warning: SSH being off now is not evidence that it was never on. Look at the change history, not the current setting.
On item 2, the advice is not uniform. ThreatDefence recommends restricting TCP 2012 as the key interim step before the fix is applied. Rescana also recommends restricting it, while warning that because Broadcom names no port, blocking 2012 alone cannot be assumed to close the flaw. Rescana further notes that the firewall in the appliance's own management interface cannot filter by port and can be switched off by anyone who reaches that interface, which means the restriction belongs on network equipment outside vCenter. Either way, a port restriction buys time. The fix is the fix.
On how far back to look, ThreatDefence advises going back to at least September 1, 2026. Mandiant, per Help Net Security and Rescana, recommends its hardening and logging tool for the vCenter appliance, egress filtering, and forwarding logs to a security monitoring system. The reason for moving logs off vCenter is plain: an attacker who is administrator can also touch the records stored on it.
6. Frequently asked questions
What is the difference between vCenter and ESXi? ESXi is the hypervisor installed on a physical server that actually runs virtual machines. vCenter is the server that manages many ESXi hosts together. If ESXi is the workforce, vCenter is the control room. The current flaw is in vCenter, and Broadcom says ESX is not affected.
If ESXi is not affected, are the virtual machines safe? No. Not vulnerable and not reachable are different things. A vCenter administrator can operate ESXi hosts and virtual machines through normal functions. In the GTIG case the attacker went through vCenter to a host and read a virtual machine's disk.
Is it fine if vCenter is not exposed to the internet? That removes the direct route from outside. But the attacker in the GTIG case entered through a flaw in a different product and then moved to vCenter. An internal-only vCenter still needs the patch.
Which version do I need? vCenter 9.1.x goes to 9.1.1.0200, 9.0.x to 9.0.2.0200, and 8.0 to 8.0 U3m. VCF 5.x needs vCenter 8.0 U3m and SDDC Manager 5.2.5 together. Because installing these can constrain later upgrade paths, read the cautions in the Broadcom advisory first.
I run vCenter 7.0. What now? There is no public fix. The guidance is that customers with an extended-support contract should contact Broadcom. Without one, you need a plan to move to a supported branch, and until then the available measure is to narrow which networks can reach vCenter down to the management network. Do not trust a scanner that marks 7.0 as unaffected.
Once I have patched, am I done? No. If the report of exploitation since early September is right, you need to establish whether anyone got in during the two-plus months before the patch. If there are signs of compromise, the signing key and stored credentials have to be rotated. A patch does not invalidate keys that have already left.
Broadcom did not say it was exploited. Can I rely on that? It is true that the advisory does not mention exploitation. But the reporter it credits, Mandiant, is part of Google, and Google GTIG reports exploitation. There is no basis for reading the advisory's silence as confirmation that nothing happened. When setting priorities, it is safer to give weight to the exploitation report.
It is not on CISA KEV. Does that make it less urgent? KEV additions lag. The flaw was simply not listed as of November 20, 2026. A 9.8 unauthenticated flaw in a management server, with a major threat intelligence team reporting real use, belongs at the front of the queue regardless of KEV.
I applied the July vCenter patch. Do I have to patch again? Yes. CVE-2026-59309 from the July 29 advisory VMSA-2026-0006 and the current CVE-2026-123890 are in the same directory service, but Broadcom says they are separate issues. According to GTIG, the attacker's tool worked against a vCenter that already had the July fix.
Is there public exploit code? None was confirmed as of November 20, 2026, and the attacker's tool has not been released. With a fix now public, though, it is reasonable to expect attempts to work backward from it to the flaw.
7. What remains open
| Item | Confirmed | Not confirmed |
|---|---|---|
| Fixed versions | 9.1.1.0200, 9.0.2.0200, 8.0 U3m (sources agree) | We did not open the Broadcom advisory itself |
| Build numbers | Numbers for the 9.1 and 9.0 branches (Rescana) | 8.0 U3m build 25934264 appears in Rescana only |
| 7.0 and 6.x | No public fix for 7.0 | Whether one will come; 6.5 and 6.7 are only presumed affected |
| Mechanism | An authentication bypass in vmdir that leaks a session key during the handshake | Three different root-cause descriptions; Broadcom gives no detail |
| Port | TCP 2012 per GTIG | Broadcom names no port; whether blocking 2012 alone is enough |
| Exploitation | The GTIG report (since early September 2026) | Confirmation from Broadcom; KEV listing. The SSVC value says none |
| Attribution | GTIG names ARMORY CASTLE (formerly UNC5221), suspected China-nexus | It is an assessment; whether other groups have the flaw |
| Victims | That there was at least one case | Number of victims, countries, sectors |
| Exposure | — | Count of internet-exposed vCenter servers; absent from all four sources |
| Public exploit code | None as of November 20 | Whether one appears later |
To sum up: vCenter does not run virtual servers. It drives the hosts that run them, and it contains the identity system that decides who may drive it. CVE-2026-123890 lets an attacker skip that identity check. The order of work is this. Upgrade to 9.1.1.0200, 9.0.2.0200 or 8.0 U3m. On 7.0, contact Broadcom or plan a migration, and narrow network access in the meantime. Then go through administrator accounts, SSH setting changes and host login records from early September 2026 onward, and rotate keys and credentials if anything turns up.
For attacks that encrypt the ESXi layer directly, see our October 12, 2026 piece, "What hypervisor ransomware is — lock the layer below and hundreds of servers stop at once." For why unpatched flaws are so valuable, see the August 10, 2026 piece, "Zero-day explained — why an unpatched flaw is the most expensive thing in security." For what happens after a fix is published, see the November 18, 2026 piece, "What is patch diffing — why a fix is a map, 63 days to −7." For why out-of-support branches get no fix, see the October 26, 2026 piece, "What is end of life (EOL) software? Why PHP2Shell fixes stopped at 8.2 — and 8.2 ends December 31, 2026."
Sources
- Help Net Security — VMware vCenter zero-day exploited for months before patch (CVE-2026-123890) (2026-11-20)
- Rescana — VMware vCenter Auth Bypass Exploited per Google (CVE-2026-123890): Patch Now (2026-11-20, data as of 11-19)
- ThreatDefence — vCenter CVE-2026-123890 Zero-Day: Patching and Hunting (2026-11-20)
- SOC Prime — CVE-2026-123890: Critical VMware vCenter Zero-Day Exploited by China-Linked Hackers (2026-11-20)