P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

Mandiant says vCenter CVE-2026-123890 was exploited 10 to 11 weeks before Broadcom's advisory, which is silent

Mandiant says a suspected China-nexus actor used VMware vCenter flaw CVE-2026-123890 in early September 2026. Broadcom's advisory of 19 November prints nothing about exploitation: 10 to 11 weeks after, by this site's count, and with no public patch stated for vCenter 7.0.

By · LinkedIn · · 22 min read

Mandiant puts use of the flaw 10 to 11 weeks before the advisory; the advisory never says so (derived)

Broadcom published advisory VMSA-2026-0009 (notification 39495) on 19 November 2026. The page prints Last Updated 19 November 2026, status OPEN, severity Critical, CVSS base score 9.8 and workaround None, and its change log has one entry: the initial advisory. The same day Google Threat Intelligence Group and Mandiant published "ARMORY CASTLE Exploits vSphere", which says the suspected PRC-nexus actor ARMORY CASTLE exploited CVE-2026-123890 "in early September 2026". The post prints no day. Counting from 1 to 10 September to the advisory date gives 70 to 79 days, which is 10.0 to 11.3 weeks (derived: 19 November minus 10 September, and minus 1 September).

Mandiant adds a second date. The Go binary it recovered carries a build timestamp of 2026-07-13T09:21:24Z. That is 129 days before the advisory (derived) and 16 days before 29 July, the day CVE-2026-59309, a different vCenter authentication bypass, was disclosed. The 16 is Mandiant's own figure and this site checked it. Mandiant also says "Go build metadata can be modified prior to compilation", so the timestamp is evidence, not proof.

Neither Broadcom page says the flaw was used by anyone. That is the checkable fact. Here is what it does not establish:

  • When use began or ended. "Early September" is Mandiant's phrase, from logs. The page names no day, and no page says whether use continued.
  • How many vCenter servers or victims. The Mandiant page says it observed the actor target an internal vCenter server. SecurityWeek says a Mandiant executive said on LinkedIn that the company investigated multiple incidents. This site did not read that post.
  • That anyone else used the flaw. No saved page says so.
  • Why Broadcom's pages are silent. No saved page gives a reason.
  • When Mandiant reported the flaw to Broadcom. Broadcom thanks Mandiant for reporting it. No page gives a date.
  • How the actor learned of the flaw. Mandiant says the actor had "knowledge" of it. It does not say the actor found it by its own research.
  • A fix for vCenter 7.0. Broadcom's matrix points 7.0 customers to extended support. SecurityWeek says there is no public patch.
  • That the other four CVEs in the advisory are exploited. No saved page says so, or says they are not.

Who says what, line by line

Diagram 1 puts the four saved pages side by side. Two of them are Broadcom's: the advisory and a questions and answers document that Broadcom publishes on GitHub. This site read every line of both for any statement about exploitation.

Drawn from the four pages saved on 19 November 2026. Wording is the pages own, shortened.

The advisory. The word "exploit" appears five times and every time inside a "Known Attack Vectors" line, in conditional form: for CVE-2026-123890, "A malicious actor with network access to vCenter may exploit this issue to bypass authentication and gain unauthorized access to the system." The introduction says the vulnerabilities "were privately reported to Broadcom". The only mention of Mandiant is the acknowledgment: "Broadcom would like to thank Mandiant for reporting this issue to us." The page does not contain "in the wild", "zero-day", "exploited" or any actor name.

The Q&A. It answers 43 numbered questions. None asks about, or answers on, exploitation or an actor. Question 2 calls the vCenter issue "an emergency change" under ITIL methodologies. Question 7 says CVE-2026-123890 "is a new and distinct issue" from CVE-2026-59309. Question 19 says there are no workarounds "as listed by Broadcom". Question 41 says all the CVEs "are attributable to specific security researchers". The word "exploited" appears once, in question 35, about a different flaw: it describes how an attacker with a local privileged process would use CVE-2026-123894, not a sighting.

SecurityWeek's claim holds. SecurityWeek writes that Broadcom's advisory "does not mention in-the-wild exploitation". This site read the whole advisory and the whole Q&A as saved and found no such statement in either. The claim describes the advisory correctly; it does not say why.

Mandiant's post says the actor exploited the flaw "in early September 2026", "according to forensic evidence", and attributes the activity to the suspected PRC-nexus actor ARMORY CASTLE, formerly tracked as UNC5221. SecurityWeek, which says it relays Google, adds that the actor is "also known as Warp Panda" and that exploitation ran "for more than two months before Broadcom released a patch". That matches the derived range: more than two months is more than 60 days, and the range is 70 to 79.

The dates, to scale, and what each one is

Diagram 2 places every date on one axis from 13 July to 19 November. Five of the six positions are dates printed on a saved page. The sixth, 29 August, is BeyondTrust's first known exploitation of its own flaw, CVE-2026-102764, taken from this site's earlier BeyondTrust briefing; this site did not fetch BeyondTrust's pages again today. Mandiant's saved page does not say whether the foothold in this intrusion was that exploitation, so the marker is context and not a link in the chain.

Drawn from the saved pages of 19 November 2026 and this site's BeyondTrust briefing. Positions come from one time scale; the day counts for September and the 129 are derived.

Table 1. Dates on the saved pages, who prints each, and the intervals worked out from them (derived unless marked).

DateWhat it is, and who prints itInterval to the advisory
13 Jul 2026, 09:21 UTCBuild timestamp in the recovered Go binary. Mandiant, with the caveat that Go build metadata can be modified129 days
29 Jul 2026CVE-2026-59309, a different vCenter authentication bypass, disclosed. Mandiant113 days
29 Aug 2026BeyondTrust first known exploitation of CVE-2026-102764. This site's earlier briefing; not Mandiant's page82 days
Early Sep 2026, no dayBONESAW run against a vCenter already patched for CVE-2026-59309. Mandiant, from logs and forensic evidence70 to 79 days (1 to 10 Sep)
15 Oct 2026Mandiant and GTIG publish their BeyondTrust post. Mandiant35 days
19 Nov 2026Advisory VMSA-2026-0009 published and last updated; the Mandiant post and SecurityWeek the same day. All four pages0

Four intervals matter. The build timestamp sits 16 days before the CVE-2026-59309 disclosure (Mandiant's figure, checked: 13 July to 29 July). It sits 50 to 59 days before the September band (derived), which fits Mandiant's remark that the build time "predates its deployment in this intrusion by over a month". The band sits 70 to 79 days before the advisory (derived). And 29 July, when the July flaw was disclosed, is 113 days before the advisory (derived).

Mandiant prints "early September" and no day, so the 10.0 to 11.3 weeks above is a range built on this site's assumption that early means the first ten days. If Mandiant meant a different span, the numbers move.

Which flaw: CVE-2026-123890 against the July CVE-2026-59309

Both are authentication bypasses in the vCenter directory service, disclosed within four months of each other, and both let an attacker obtain a shared session key. Mandiant says so and then lists differences. Broadcom's Q&A says only that CVE-2026-123890 "is a new and distinct issue". Table 2 sets the two descriptions side by side in the post's own terms. This site copies no detail of how either is triggered.

Table 2. The two flaws as the Mandiant post describes them, with Broadcom's wording for the new one. Read 19 November 2026.

PointCVE-2026-59309 (disclosed 29 July)CVE-2026-123890 (advisory 19 November)
ComponentThe vmdir daemon of vCenter ServerThe vmdir DCE/RPC interface (Broadcom: the VMware Directory Service)
Protocol and reachLDAP, to TCP ports 389 or 636, by a remote unauthenticated attacker with network accessDCE/RPC; the post names an unauthenticated Secure Remote Password (SRP-6a) session key disclosure
ResultBypasses authentication to forge administrative sessions and create rogue SSO accountsA remote authentication bypass giving an authenticated administrator session (Mandiant); unauthorized access (Broadcom)
What they shareObtaining the shared session key "K" without knowing the passwordObtaining the shared session key "K" without knowing the password
Broadcom on the linkQ7 refers to it as CVE-2026-59309 from VMSA-2026-0006Q7: "a new and distinct issue"; all systems below the listed versions need this update

The two descriptions agree that the flaws are different and disagree on nothing a saved page lets this site test. The post also names a cause for CVE-2026-123890 in the same paragraph. This site does not repeat it, because a cause is not needed to patch or to hunt.

The date of knowledge, and who knew what before the patch

Mandiant's claim is narrow: the actor "had knowledge of CVE-2026-123890 prior to the disclosure of CVE-2026-59309 on July 29, 2026". It rests on two pieces of evidence, and the post prints the weakness of the second.

For the claim. Logs show BONESAW was run in early September against a vCenter server that was already patched against CVE-2026-59309. If the logs are right, that shows the tool did not depend on the July flaw. And the recovered binary carries a build timestamp 16 days before the July disclosure.

Against reading too much into it. The post itself says "although Go build metadata can be modified prior to compilation". The first piece of evidence shows only that the tool used a different flaw from the July one; on its own it does not date the actor's knowledge. Mandiant's reading, in its outlook section, is that BONESAW "was likely developed for broader operations". That is Mandiant's inference, and so is the step from a timestamp to knowledge.

Credit. Broadcom thanks Mandiant for reporting the flaw, and SecurityWeek says "Mandiant, which reported the bug to Broadcom". Mandiant says the actor used it in September. Put together, Mandiant knew of the flaw before the patch existed, since the patch followed its report. The pages do not say when Mandiant first saw the use, whether it found the use before or after reporting the flaw, or when it reported. So that Mandiant knew of the exploitation before the patch is a reasonable guess, and not a statement on any saved page.

Versions: the matrix, vCenter 7.0, and what Mandiant says the tool runs on

The advisory's response matrix for CVE-2026-123890 has seven rows, all rated 9.8 Critical, all with workarounds None. Table 3 copies them. The Q&A says to install a fixed version, and that the patches are cumulative, so later versions also resolve the issues.

Table 3. Response matrix rows for CVE-2026-123890, as printed in advisory VMSA-2026-0009 (Last Updated 19 November 2026, saved about 20:26 UTC).

Product and componentVersionFixed version as printed
VMware Cloud Foundation, VMware vSphere Foundation: vCenter9.1.x.x9.1.1.0200
VMware Cloud Foundation, VMware vSphere Foundation: vCenter9.0.x.x9.0.2.0200
VMware vCenter8.08.0 U3m
VMware vCenter7.0Contact Broadcom Support if you have extended support contract.
VMware Cloud Foundation: vCenter5.x5.2.5 (Async Patching Guide: KB88287)
VMware Telco Cloud Platform: vCenter3.0, 4.x, 5.0.x, 5.1.x, 5.2KB463116
VMware Telco Cloud Infrastructure: vCenter3.0KB463116

vCenter 7.0. The matrix lists it: one row, no fixed version, a pointer to Broadcom Support for customers with an extended support contract. Q31 says "Yes", 7.0 is affected, "and the advisory has been updated", notes that vSphere 7 reached End of General Support on 2 October 2025 (413 days before the advisory, derived), and tells customers with an active extended support contract to "use those support processes to request patches". SecurityWeek says "there is no public patch for version 7.0". The three agree: a patch may exist for some customers, and no page prints a public one. One detail does not line up: Q31 says the advisory has been updated, while the saved advisory's change log has only the initial entry and its updated line reads "2026-11-19 (Initial Advisory)". The matrix row for 7.0 is present in the saved copy, so the update may have been made before publication; no page says.

Older versions. Q30 says to presume that vSphere 6.5 and 6.7 are affected, because Broadcom does not evaluate products past End of General Support. Neither is in the matrix.

What Mandiant says the tool runs on. The post says BONESAW "achieves full system compromise against vCenter Server 7.x and 8.x appliances". Broadcom's matrix also lists 9.1, 9.0, VMware Cloud Foundation 5.x and Telco Cloud rows. The post does not say whether the tool works on 9.x, and Broadcom does not say whether 9.x was reached; Broadcom's matrix says 9.x is affected and needs a fix. Read together, the evidence of use is on 7.x and 8.x and the affected list is wider.

Two small differences. The matrix gives 5.2.5 as the fixed version in the VMware Cloud Foundation 5.x vCenter row, with the async patching guide; Q17 says Cloud Foundation 5.x customers apply vCenter 8.0 U3m through that process and update SDDC Manager to 5.2.5. This site does not reconcile the two. And Q10 says ESX is not affected: the flaw is in vCenter, not the hypervisor, although SecurityWeek relays that the actor later reached an ESXi host.

Operational notes from the Q&A, for planning. The vCenter fixes are not eligible for vCenter Quick Patch (Q27). Applying 8.0 U3m or 9.0.2.0200 blocks upgrades to Cloud Foundation 9.1.1 until 9.1.2 (Q39, "Back-in-Time"). Running workloads continue during a vCenter update (Q25). Enhanced Linked Mode does not protect (Q22).

Severity: one flaw scored 9.8, one of five reported used

The advisory header prints "Severity: CRITICAL", "CVSS Base Score 9.8" and "CVSSv3 Range: 6.7-9.8". The 9.8 is the vCenter flaw. The range is across five CVEs in three products. The advisory's reference list links the FIRST calculator with a vector for each CVE; for CVE-2026-123890 the vector in that link is AV:N, AC:L, PR:N, UI:N, S:U, C:H, I:H, A:H (separated by commas here, with slashes in the link), which is network, low complexity, no privileges, no user interaction, with high impact on confidentiality, integrity and availability. The scores are Broadcom's. This site did not check the CVE records, because it fetched nothing beyond the four saved pages.

Table 4. The five CVEs in VMSA-2026-0009: Broadcom's score and word, and what the saved pages say about use.

CVE and flaw as Broadcom names itScore and Broadcom's wordReported exploited on the saved pages
CVE-2026-123890, vCenter authentication bypass in the VMware Directory Service9.8, CriticalYes, by Mandiant (early September 2026). Not by Broadcom
CVE-2026-123891, SDDC Manager SQL injection (authenticated actor)8.1, ImportantNo saved page says either way
CVE-2026-123892, SDDC Manager authenticated command injection7.2, ImportantNo saved page says either way
CVE-2026-123893, SDDC Manager local privilege escalation (administrative privileges)6.7, ModerateNo saved page says either way
CVE-2026-123894, VMware Tools for Windows SAML token signature bypass (local privileged process, guest alias mapping)6.7, ModerateNo saved page says either way

A reader skimming the header sees one score, 9.8, and one word, Critical, and could take it for the whole advisory. The other four flaws need an authenticated actor or local privileges. Only CVE-2026-123890 is reported used, and the report comes from the finder, not the vendor.

The chain as Mandiant tells it, and where SecurityWeek adds to it

The Mandiant post continues a story from 15 October, when it published its BeyondTrust account: a suspected PRC-nexus actor exploiting CVE-2026-102764, a Remote Support zero-day patched in BT26-05. This site covered the vendor side in its BeyondTrust briefing. The 19 November post says "In this campaign, ARMORY CASTLE exploited two zero-days, and deployed three sets of malware specifically tailored for operating on targeted devices", which Mandiant reads as "deep platform expertise and substantial resource backing".

Diagram 3 draws the sequence at the level the pages allow, with the source of each step. Mandiant's saved page confirms the BeyondTrust zero-day as part of the campaign, the vCenter step in early September, and a further Linux appliance. SecurityWeek says the actor gained initial access through the BeyondTrust zero-day, pivoted to Linux appliances on the internal network, targeted vCenter in early September, forged SAML tokens, installed backdoors, took the vCenter master encryption key and database credentials, and moved to an ESXi host. Those last items are in SecurityWeek's relay and not in the saved Mandiant copy, which lacks the step by step intrusion description.

Drawn from the Mandiant post and the SecurityWeek article as saved on 19 November 2026. No exploit step is drawn.

What the Mandiant page says BONESAW does with the flaw, at the level it prints: it copies the directory database (an LMDB file), takes the signing key of the vCenter token service, and creates a single sign on administrator account; the page also says it can create a backdoor machine account and forge administrative SAML tokens. On the additional Linux appliance, Mandiant describes a Slack based backdoor it calls LAZYBONES, with a command channel over HTTPS. Mandiant says it saw the same backdoor in a separate campaign in early 2025 after a different flaw, CVE-2025-22457.

The attribution paragraph is short: GTIG "attributes this activity to the suspected PRC-nexus espionage actor ARMORY CASTLE", based on tooling and techniques that "align with previously observed activity", and for LAZYBONES "directly overlap". That is a vendor attribution, labelled suspected, and nobody independent has confirmed it on any saved page.

No victim is named on any saved page. This site names none and does not guess.

Labels that read like evidence

A label is not a control and not an observation. Six appear in the pages, each read against what the page prints.

Table 5. Labels in the saved pages, what each establishes and what it leaves open.

Label and whereWhat it establishesWhat it does not
"Known Attack Vectors" (advisory heading)What an attacker with network access could do, in conditional wording ("may exploit")That anyone has. The heading reads like a list of observed attacks
"Privately reported" (advisory introduction)The flaw reached Broadcom before the patchWhether it was used before the report
"Workarounds: None" (advisory, every row)Broadcom lists no workaround. Q19 says other controls may exist and each organisation decidesThat no mitigation helps. Mandiant recommends hardening and a firewall posture, which is not a listed workaround
"Zero-day" (SecurityWeek; Mandiant for the campaign)By Mandiant's account the flaw was used before a patch existedAnything in Broadcom's pages: neither uses the word
"Suspected PRC-nexus" (Mandiant and GTIG)GTIG's view of who is behind itProof, independence, or a victim count
"Source of truth" (Q&A, about the advisory)Which products and versions are affected, and the patchesWhether exploitation has been seen

What the pages state, and what they leave open

Table 6. What the four saved pages establish and what they leave open.

QuestionStatedNot stated
Was CVE-2026-123890 exploited?Mandiant: yes, early September 2026, by ARMORY CASTLE. SecurityWeek relays itBroadcom: nothing in the advisory or Q&A
By whom?GTIG: the suspected PRC-nexus actor ARMORY CASTLE, formerly UNC5221Independent confirmation; any other actor
How many victims or vCenter servers?Mandiant observed the actor target an internal vCenter server. SecurityWeek: multiple incidents, per a LinkedIn post this site did not readA count, names, sectors, countries
When did it start?Early September 2026; build timestamp 13 July 2026The day; whether the build timestamp was altered; when the actor learned of the flaw
Which versions?Broadcom: 9.1, 9.0, 8.0, 7.0, Cloud Foundation 5.x, Telco Cloud. Mandiant: the tool works on 7.x and 8.xWhether 9.x was reached; a public fix for 7.0
Which fix?9.1.1.0200, 9.0.2.0200, 8.0 U3m, 5.2.5, KB463116A workaround
Who reported it, and when?Broadcom: Mandiant reported itThe date; whether Mandiant saw the use first
Are the other four CVEs used?NothingEither way

Where the pages differ, with nothing resolved

Table 7. Differences between and inside the saved pages. Each is reported as printed; none is resolved.

PointOne page printsAnother prints
ExploitationMandiant: used in early September 2026Broadcom advisory and Q&A: nothing; the advisory only says "may exploit" in a conditional line
Does the advisory mention it?SecurityWeek: it "does not mention in-the-wild exploitation"This site read the advisory and Q&A: that holds for both
Initial accessMandiant: the campaign used two zero-days and follows its 15 October BeyondTrust postSecurityWeek: "the hackers gained initial access through CVE-2026-102764". The saved Mandiant copy has no sentence saying so for this victim
Which flawBroadcom Q7: new and distinct from CVE-2026-59309Mandiant: shares "some overlap", the shared session key "K", with multiple key differences
VersionsMandiant: the tool works on vCenter 7.x and 8.xBroadcom matrix: fixed versions for 9.1, 9.0 and 8.0; a contact line for 7.0. SecurityWeek: no public patch for 7.0
SeverityBroadcom: 9.8 for the vCenter flaw, a range of 6.7 to 9.8 across fiveSecurityWeek repeats 9.8; the Mandiant post prints no score
Date of knowledgeMandiant: it has evidence of knowledge before 29 JulyMandiant: "Go build metadata can be modified"
CreditBroadcom: thanks Mandiant for reportingMandiant: the actor used it in September. Neither prints when Mandiant reported
Stolen materialMandiant (saved copy): the token service signing keySecurityWeek: also the vCenter master encryption key and database credentials
Whether the advisory was updatedQ&A Q31: "the advisory has been updated"Saved advisory: change log has one entry, the initial advisory
The actor's other namesMandiant: ARMORY CASTLE, formerly UNC5221SecurityWeek: also known as Warp Panda

What the pages print for a defender

This section copies the defensive items the saved pages print, as printed and attributed. It prints no command line, request, input or code of the flaw, and no network address: the Mandiant post lists 11 network indicators, described as suspected obfuscation network addresses, and this site prints none of them. The saved copy of the post also leaves out the intrusion description, the exploit tool's arguments, descriptions of the further tools, and the YARA rules; this site prints none of that either, and the post itself is the place to read it. The post says its indicators are also in a GTI Collection for registered users; this site did not read that collection, the defender's guide it recommends, or the hardening tool.

The first block is the advisory's own header and its entry for the vCenter flaw: the severity, the one line of remediation, and the score vector from the calculator link in its reference list. Broadcom prints the vector inside a link; this site shows the vector only.

Text
Advisory ID: VMSA-2026-0009
Advisory Severity: Critical
CVSSv3 Range: 6.7-9.8
Notification Id: 39495
Last Updated: 19 November 2026
Status: OPEN
CVSS Base Score: 9.8
WorkAround: None
3a. vCenter authentication-bypass vulnerability (CVE-2026-123890)
Resolution: To remediate CVE-2026-123890 apply the patches listed in the 'Fixed Version' column of the 'Response Matrix' found below.
Workarounds: None
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
From Broadcom advisory VMSA-2026-0009, notification 39495, as read on 19 November 2026 (saved about 20:26 UTC).

The second block is what Mandiant prints for restricting the management interfaces: row headings and one sentence from its hardening table, and the port of the appliance management interface. The post says the exploit tool reaches the directory service over port 2012. The heading of the third row prints 2010 where its sentence prints 2012; this site reports both as printed and does not resolve it. A defender decides which hosts may reach which ports. The post warns that iptables rules need rigorous testing and formal change control before production.

Text
VCSA iptables Inbound DROP Policy (TCP 443)
VCSA iptables Inbound DROP Policy (TCP 2010)
An iptables DROP policy on TCP 2012 for non-allowlisted IPs prevents the exploit payload originating from a compromised Linux appliance from reaching the vmdird service.
VCSA iptables Inbound DROP Policy (TCP 22 & 5480)
VCSA iptables Egress DROP Policy
the VAMI, which resides on TCP port 5480
From the Mandiant and GTIG post, hardening section and Table 1, as read on 19 November 2026.

The third block is the post's file indicator table, copied as printed: a filename or path and the post's name for it. The page prints no hash values, so there are none to copy.

Text
ActiveSer.jsp  SLIMSAVAGE Webshell
u.jsp  CAGEMATCH Staging Webshell
/tmp/123  BONESAW Utility
/tmp/sss  SAWSNAKE Utility
/tmp/dbr  BEARBONES Backdoor
/usr/bin/vmtoolscs  LAZYBONES Backdoor
From the Mandiant and GTIG post, Table 3 file indicators, as read on 19 November 2026. The page lists no hashes.

The fourth block is the names of the Google Security Operations rules the post says detect the activity. Rule logic is not printed on the saved copy.

Text
VMware vCenter VAMI API SSH Enablement
Base64 Encoded Shell Command Execution
Linux Web Server Process Echo Base64 Encoded String
Suspicious Curl Network Connection to Staging Directory
Suspicious Curl Download to tmp
From the Mandiant and GTIG post, Google SecOps section, as read on 19 November 2026.

The fifth block is how Broadcom's Q&A says to read a vCenter build and version from PowerCLI, the check a defender needs to compare against Table 3.

Text
Connect-VIServer
$global:DefaultVIServer.Build
$global:DefaultVIServer.Version
From Broadcom's Q&A, question 23, as read on 19 November 2026.

What to do, in order

First, what Broadcom says: apply the fixed version for your branch from the response matrix; Broadcom lists no workaround. The rest is this site's suggestion, drawn from what Mandiant and SecurityWeek print and labelled where it goes beyond them.

Take this with you

A checklist for a vCenter owner

  • Broadcom: apply the fixed version in the response matrix (9.1.1.0200, 9.0.2.0200, 8.0 U3m, 5.2.5 with KB88287 for Cloud Foundation 5.x, KB463116 for Telco Cloud). Broadcom lists no workaround.
  • This site's suggestion: inventory every vCenter, including those inside VMware Cloud Foundation and Telco Cloud and any a supplier runs for you, and record each version. Broadcom's Q&A shows how to read the build and version from PowerCLI.
  • This site's suggestion: patch to the fixed version for the branch. Where the branch is 7.0, Broadcom says to use the extended support process; no page prints a free fix. Plan the upgrade the Q&A urges (Cloud Foundation 9.1) and read the Back-in-Time note if an upgrade is in progress.
  • Mandiant and SecurityWeek, via Mandiant: patching is not enough. Examine the system for evidence of compromise, and for use before 19 November, since Mandiant says use began in early September. This site's suggestion for what to look at, from what the post says the tool does: new single sign on administrator accounts, new machine accounts in the directory, signs that the token service signing key was copied, and the Google SecOps rule names printed above.
  • Mandiant: restrict the management interfaces and ports to named administrator hosts, with a default deny posture and logging of refused connections, and test any firewall change under change control.
  • This site's suggestion: if you run BeyondTrust Remote Support, treat that appliance as a possible way in. Mandiant says the campaign used the BeyondTrust zero-day CVE-2026-102764, and SecurityWeek says it was the initial access; the vendor side is in the earlier briefing.
  • This site's suggestion: after patching and a compromise check, plan to rotate what a vCenter compromise exposes. SecurityWeek relays that the actor took the token signing material, the master encryption key and database credentials. Mandiant's saved text names only the signing key.
  • Mandiant: control outbound traffic from the vCenter appliance. The post says a default deny egress policy stops a backdoor reaching an external command channel.

The question the silence leaves open

Broadcom's advisory says how to fix the flaw and what the flaw lets an attacker do. It does not say whether anyone had. The finder says someone had, from early September, and the finder is a vendor with a view of its own. If the vendor advisory says nothing about exploitation and the finder says September, which date starts your investigation?

Key facts

Sources

  1. PrimarySupport content notification 39495, advisory VMSA-2026-0009: header, the five CVE entries, the response matrix, scores, acknowledgments and change log. Last Updated 19 November 2026; saved about 20:26 UTC on 19 November 2026Broadcomaccessed 2026-11-19
  2. PrimaryVMSA-2026-0009: Questions and Answers (43 questions), published by Broadcom on GitHub; read for vCenter 7.0 (question 31), workarounds, the link to CVE-2026-59309 and upgrade notes. Saved about 20:26 UTC on 19 November 2026Broadcomaccessed 2026-11-19
  3. PrimaryARMORY CASTLE Exploits vSphere, 19 November 2026: the vendor's exploitation claim, attribution, evidence of prior knowledge, hardening advice and file indicators. The saved copy omits the intrusion description, command lines, 11 network indicators and YARA rulesGoogle Threat Intelligence Group and Mandiantaccessed 2026-11-19
  4. Reported byNewly Patched VMware vCenter Flaw Exploited as Zero-Day by Chinese APT, published 19 November 2026 at 19:21 UTC per the page data. Secondary: relays Mandiant and states that Broadcom's advisory does not mention exploitationSecurityWeekaccessed 2026-11-19

Share this briefing

Know someone who owns this problem? Send it to them.

Related briefings

The briefing, in your inbox

Practitioner analysis of cyber and AI security news. No vendor noise.

How often

Every new briefing in one email, at 7am, or at 7am, 12:30pm and 6pm. Nothing is sent when nothing is new. Unsubscribe any time.