Broadcom patched a critical vCenter authentication bypass without offering a workaround. Google documented exploitation in early September, making a compromise review relevant alongside patching.
On a vCenter server already patched for July's authentication flaw, BONESAW ran in early September 2026. Google's investigators traced it to a different vulnerability, CVE-2026-123890. Broadcom's critical advisory arrived on 19 November 2026 with a CVSS 9.8 score, patches issued and no workaround. An update on 20 November 2026 clarified which VMware Tools versions were affected. A score describes the opening; the elapsed time determines how much investigation the owner needs to buy.
We would pay for an emergency patch and a compromise review, while holding Broadcom shares rather than selling on the severity score alone. We also expect CISA to list CVE-2026-123890 in its Known Exploited Vulnerabilities catalog by 31 December 2026. We call this Evidence Escalation: technical severity dictates patch urgency, while independently documented exploitation supplies the case for federal known exploited status. Our prediction concerns the catalog entry, rather than the number of organizations affected. That distinction determines what an owner needs to fund and what an investor should price.
Broadcom's notice addresses five vulnerabilities across vCenter, SDDC Manager and VMware Tools, with scores from 6.7 to 9.8. The 9.8 belongs to an authentication bypass in the VMware Directory Service. Network access can be enough to gain unauthorized access, according to the advisory. Fixed vCenter versions include 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. Broadcom's Q&A calls the fix an "emergency change." That gives administrators a priority and a route to action. Rapid7 says vCenter Quick Patch will not apply these fixes; management interfaces go temporarily offline while running workloads continue. The interruption is limited, making the cost of postponement harder to defend.
Google's Mandiant researchers supply the uncomfortable extra detail: "ARMORY CASTLE exploited this vulnerability in early September 2026 using BONESAW, according to forensic evidence." Google attributes the campaign to a suspected China nexus actor. Exploitation created an administrator session; BONESAW copied the vCenter directory database and took the SAML token signing key. SAWSNAKE used stolen signing material to forge tokens that bypass multifactor authentication. The consequence extends beyond the entry point because the attacker has acquired a way to impersonate trusted identities.
A fix closes the authentication bypass; a compromise review asks what happened while the bypass was open. That interval matters more than a decimal score. Counting from 1 September 2026 to the 19 November 2026 patch gives 79 days; counting from 10 September gives 70. Google dates the observed activity to early September, so 70 to 79 days is the useful bound. The window describes the lead over the patch, and it is long enough to change a maintenance decision into an investigation budget.
Broadcom credits Mandiant for finding CVE-2026-123890, yet its advisory describes a potential attack vector without mentioning actual exploitation. Canada's Cyber Centre took a different approach on 19 November 2026: "Open source reporting indicates that CVE-2026-123890 is exploited in the wild." Those documents serve different audiences. One tells operators what to install; the other supplies a reason to inspect what the attacker might already control. Spending only on installation buys the easier half of recovery.
The strongest objection comes from Google's own investigation. ARMORY CASTLE first entered through a separate zero day in BeyondTrust Remote Support, then moved through internal Linux appliances without endpoint detection and response tooling. Google describes two zero days and three malware sets in this targeted campaign. The path required more than spotting an exposed vCenter: it involved a working foothold and an environment with detection gaps. Broadcom's Q&A says "the specific response timing depends on your unique circumstances." That is a defensible reason to tailor the investigation rather than assume a universal breach.
The targeted reading has to account for what happened during that very campaign. Google documented the exploit 70 to 79 days before the November patch, on a vCenter installation that had received July's separate fix. Its investigators also documented stolen signing material and forged authentication tokens, whose risk remains after the vulnerable entry point is repaired. A compromise review can look for those consequences without presuming that every vCenter has been breached. That is a much more exacting job than reading 9.8 and scheduling a restart, and it directs security spending toward the right machines.
There is an equally serious objection to our federal prediction. In July, Broadcom disclosed another vCenter Directory Service authentication bypass, CVE-2026-59309, also rated 9.8. It remained absent from CISA's catalog in the 18 November 2026 release, 112 days after the 29 July 2026 disclosure. Severity alone had bought that vulnerability no federal designation. The precedent is particularly awkward for anyone arguing that a 9.8 must inevitably turn into a KEV entry. We are arguing for a listing because of the exploitation evidence instead.
The same July advisory supplies the other half of the comparison. Its CVE-2026-59310, a vCenter syslog directory traversal also rated 9.8, entered the KEV catalog on 18 August 2026, 20 days after disclosure, with ransomware use marked "Known." That was a different flaw and a different sort of campaign. November has its own evidence, from Google and Canada's Cyber Centre, that the July unlisted bypass lacked at disclosure. Evidence Escalation is a prediction about how CISA will distinguish documented use from technical possibility. A listing would impose a federal remediation deadline, increasing the cost of waiting.
The investment calculation is less dramatic. Broadcom's fiscal third quarter, ended 2 August 2026, produced $29,591 million in revenue, including $8,752 million in infrastructure software, as well as $13.7 billion in free cash flow and $24.0 billion in cash. In a scenario assigning the entire infrastructure software segment to the risk, $8,752 million over $29,591 million is 29.6% of quarterly revenue. VMware sits inside that segment, so this scenario describes revenue scope rather than an expected loss. Those numbers argue for paying for remediation without pretending the security advisory prices Broadcom's equity.
There is one group for whom the cost of waiting is particularly tangible. The November advisory gives vCenter 7.0 owners a support instruction: "Contact Broadcom Support if you have extended support contract." July's critical vCenter advisory added guidance for 7.0 on 19 August 2026, 21 days after its publication. For these customers, entitlement and vendor response become part of incident planning. A federal catalog entry would add a remediation clock to an already awkward commercial dependency.
We would keep Broadcom shares on hold while funding patches and checks for stolen vCenter credentials and signing keys, and we expect the documented September use to earn CVE-2026-123890 a CISA catalog entry. We would be wrong if the CISA Known Exploited Vulnerabilities catalog contains no entry for CVE-2026-123890 on 31 December 2026.
