Commzgear Weekly — Week of Nov 16, 2026
Nov 20, 2026 · @Jason Blackmon
Welcome back. Last week was about the tools that manage everything else. This week went one layer deeper: the hypervisor your servers run on, the firewall at your front door, the Java runtime inside half your business apps, the web server in front of all of it, and a marketing script you probably forgot was on your website. None of these are things most people look at day to day, and that’s exactly why attackers like them. Here’s what happened and what I’d actually do about it.
1. A vCenter Login Bypass Was Exploited as a Zero-Day for Two Months
On November 19, Broadcom patched CVE-2026-123890 (rated 9.8), an authentication bypass in vCenter’s directory service. Someone who can reach vCenter over the network can get an admin session without a password. Broadcom’s advisory doesn’t mention attacks, but Mandiant, which reported the bug, says a suspected China-linked group it calls Armory Castle (formerly UNC5221) was using it as a zero-day for more than two months. The group got in through a recently patched BeyondTrust Remote Support zero-day, moved to Linux appliances on the internal network, then hit vCenter in early September. From there it created its own single sign-on admin account, forged login tokens, planted backdoors, stole the vCenter master encryption key and moved on to ESXi hosts. Fixed builds are vCenter 9.1.1.0200, 9.0.2.0200 and 8.0 U3m. vCenter 7 left general support in October 2025, so there’s no public patch for it.
My take: vCenter is the keys to every virtual server you own, and this attack chain is a textbook example of why “it’s only reachable from inside” isn’t a defense. They came in through a remote support tool, then walked sideways to vCenter. Patch this week, and put the vCenter management interface on its own admin network that regular servers and user machines can’t reach (same idea as the management VLAN I keep pushing in my small business switch roundup). Then check for SSO admin accounts you didn’t create. If you’re still on vCenter 7, this is the end-of-life problem in its plainest form: a 9.8 bug, exploited in the wild, and no public fix. Either pay for extended support or put the upgrade on the calendar now.
2. Check Point Firewalls Backdoored, Then Used to Wipe VMware Hosts
Check Point says attackers exploited CVE-2026-123039, a bug that lets a logged-in user on the gateway become root, against “a couple of dozen customers” in late October. Initial access came from two other flaws tied to what’s being called PHP2Shell. Once in, the attackers installed a passive backdoor Check Point calls TSGA, which can steal configs, survive in backups and snapshots, and tamper with firewall logs. They then used SSH and VMware’s vpxuser service account to reach an ESXi host and ran a wiper that zeroes out and deletes files. Affected versions are R81 through R82.20 across gateways, management and log servers, Multi-Domain and Maestro. A fix was out about a week before the attacks. CISA added the bug to its exploited list on November 13. Check Point won’t say who did it; the toolkit resembles one tied to the same Armory Castle group, but Check Point says the evidence isn’t strong enough to name anyone.
My take: Two big stories, one pattern: get in at the edge device, then go for the hypervisor. A firewall that’s been backdoored will happily hide its own logs, so “nothing looks wrong” doesn’t tell you much. If you run Check Point, confirm you’re on the fixed build per advisory sk1000362, then look at the gateway itself, not just the traffic going through it: unexpected processes, changed configs, and any SSH from the firewall into your virtual environment. That last one should never happen. Firewalls have no business logging into ESXi, so block it, and make sure backups for your hypervisors are offline or immutable.
3. “Java4Shell”: A Bug in Java Itself Puts Jira, Jenkins and Keycloak at Risk
On November 18, researchers at Calif disclosed CVE-2026-114938, a memory overflow in Java’s fast Base64 decoder. A malformed string ending in an extra “=” makes Java write past its buffer before it rejects the input. Calif says that’s enough for unauthenticated code execution in default installs of Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate, and that list isn’t complete. Two catches: it only hits x86-64 servers with newer CPUs (Intel Ice Lake and later, AMD Zen 4 and later; ARM is safe), and so far other labs have reproduced crashes and a memory leak but not code execution. Affected: Java 17.0.6 and later, plus 21, 25 and 27. Oracle, Amazon Corretto, Azul Zulu and GraalVM have fixed builds; Red Hat, Temurin and Microsoft’s builds were still pending. Severity scores range from 8.1 (Oracle) to 9.8 (Red Hat), and no attacks have been reported.
My take: The hard part with this one isn’t the patch, it’s finding every copy of Java. It’s bundled inside apps, appliances and containers, and lots of vendors ship their own runtime. Start with anything internet-facing that takes a login: SSO, Jira, Jenkins, your artifact repo. Update the Java runtime where you control it, push your vendors for updated builds where you don’t, and until then use the JVM flag -XX:-UseBASE64Intrinsics that researchers say stops the overflow. And if your Jenkins or Jira is reachable from the internet with no VPN in front of it, this is a good week to change that.
4. A 10-Year-Old NGINX Bug Can Lead to Code Execution
Also on November 18, F5 fixed five NGINX flaws, all rated 9.2. The headliner is CVE-2026-123007, nicknamed “nginx2root,” in the HTTP/2 code that’s been in every open-source release from 1.9.5 (2015) through 1.31.8. With HTTP/2 on, an attacker can run code as the NGINX worker process, though it needs memory protections disabled or bypassed, and getting full root needs a second bug. Rapid7’s demo pushed about 128 GB of traffic in roughly 2.5 minutes, so it isn’t quiet. There’s public proof-of-concept code, but no attacks reported. Fixed versions are 1.31.9 (mainline) and 1.30.7 (stable), and NGINX Plus 37.3.0.1 and 37.0.8.1. NGINX Plus R34 to R36 had no fix listed, and Ubuntu and Debian hadn’t shipped packages yet.
My take: NGINX is everywhere: reverse proxies, load balancers, Kubernetes ingress, and inside a lot of appliances and NAS boxes you’d never think to check. Upgrade, and do a full restart, not just a reload, because a reload can leave the old code running. If you can’t upgrade yet, turning off HTTP/2 blocks this bug (but not all five). The real lesson: a 10-year-old bug means “mature and widely used” isn’t the same as “safe.” Keep an inventory of what’s running in front of your apps so you know where to look when the next one drops.
5. A Website Testing Tool Served Malware Through 178 Big-Name Sites
On Monday, November 16, attackers used a flaw in Optimizely’s Experimentation product (a tool marketing teams use to A/B test their websites) to inject malicious JavaScript into customer projects. Researchers found the code on sites belonging to 178 organizations, including CNN, the BBC, IKEA, HSBC and Salesforce, without any of those companies’ own servers being hacked. The script swapped crypto wallet addresses for the attacker’s, skimmed card numbers typed into payment forms, stole cookies and took screenshots. On iPhones and iPads it quietly loaded two known iOS exploit kits, Coruna and DarkSword. Scott Helme says crypto was actually stolen; it’s not yet clear whether any iPhones were compromised. Optimizely says the hole is fixed and the code is gone.
My take: Every third-party script on your website runs with the same trust as your own code. If your marketing person added an analytics, chat or testing tag three years ago, it’s still there, and its vendor’s security is now your security. Pull up your site’s list of third-party scripts and remove anything nobody uses. If you take payments, use your processor’s hosted payment fields (cards entered in Stripe and Adyen’s hosted fields weren’t affected here). And for everyone: keep your phone updated. The exploit kits went after older iOS versions, which is the same “boring updates are the defense” lesson as my FBI public USB charging post.
6. N-able, Round Two: Email Security Cloud Breached, N-central Now Being Exploited
N-able had another rough week. Starting November 16, its hosted Mail Assure and SpamExperts email filtering services had an outage across the Americas, Europe and APAC, and N-able later confirmed an attacker got into one server in that cloud. Mail still flows and filtering still works, but the management portal, APIs, quarantine and logs were still offline as of November 19. N-able says it has no evidence customer data was taken and that this is unrelated to last week’s N-central exploits. Separately, those N-central chains are now being used for real: CISA added an Apache Axis bug (CVE-2026-90436) to its exploited list on November 18, noting it’s being chained with the 4EverAble2Shell flaws to install crypto miners. Rapid7 also caught the same Axis bug being used against an Atlassian Confluence Data Center server, where the attacker went after credentials, the domain controller and Microsoft 365 within about 12 hours. N-central 2026.4 HF3 (2026.4.1.39) turns off the Legacy API that the chains depend on.
My take: Last week I said the people hurt first would be the shops that couldn’t patch overnight. That’s now happening, and miners are usually the first wave, not the last. If you’re on N-central, get to HF3 and treat any server that had the Legacy API exposed as possibly compromised until you’ve checked it. If you run Confluence Data Center, upgrade to 9.2.28 or 10.x or turn off the Remote API setting. And if you’re a Mail Assure or SpamExperts customer, read what N-able sent your partner account and plan for a few days without quarantine access. As I said in my Ubiquiti breach post, go by the vendor’s latest update, not the first one.
Quick Hits
- UniFi Access gets seven critical and high fixes. Ubiquiti’s November 18 bulletin covers the UniFi Access door-control app, with four bugs rated 9.9 and one that allows code execution as root. No attacks reported. Update the Access application to 4.3.24 or later (it’s the app, not the reader firmware). Door access is physical security, so keep that controller off the open internet.
- F5 BIG-IP: six critical bugs, hotfixes only. F5’s November 18 advisories cover six 9.8-rated overflows that can crash BIG-IP’s traffic engine and may allow code execution, each tied to specific iRule or APM setups. The only fixes are engineering hotfixes that F5 says skip its full QA. Check whether your configs match before you rush a hotfix into production.
- Synology NAS bug used for ransomware. CISA added CVE-2024-10441 in Synology DSM to its exploited list on November 18, linked to C3RB3R ransomware hitting NAS devices. It’s a 2024 bug, so if your NAS hasn’t updated since then, fix that today. As my home router guide says, don’t expose admin pages or file shares to the internet.
- LLMjacking via LiteLLM. The same CISA update added CVE-2026-40217 in LiteLLM, a popular AI gateway. Google says attackers turned limited AI keys into full admin access. If you self-host AI tools, they’re servers like anything else: patch them and keep them off the internet.
- Nvidia investigating ransomware claims. The Gentlemen ransomware group has listed Nvidia on its leak site with a 10-day countdown and no samples. Nvidia says a third-party partner may be involved and it sees no sign its own systems were hit. Unverified for now, but it’s another reminder that your vendors’ security is part of yours.
Final Thought
Look at the path attackers took in the two biggest stories: in through a remote support tool or a firewall, then straight to the hypervisor. They aren’t after one laptop. They want the box that runs every other box. So this weekend’s homework is about the layer underneath. Write down where your virtual servers run (vCenter, ESXi, Hyper-V, Proxmox, whatever it is), what version it’s on, and whether it’s still supported. Then answer one question honestly: can a regular server or a user machine on your network reach its management page? If the answer is yes, that’s the gap to close before the next one. (And if you have a NAS sitting in a closet that nobody’s logged into since 2024, add it to the list.)
See you next week.
— Jason
Sources
- Newly Patched VMware vCenter Flaw Exploited as Zero-Day by Chinese APT — SecurityWeek, Nov 19, 2026
- Critical VMware vCenter Flaw (CVSS 9.8) Lets Network Attackers Bypass Authentication — The Hacker News, Nov 19, 2026
- Hackers Exploit PHP2Shell Flaws in Check Point Gateways to Deploy TSGA Backdoor and Wipe ESXi Files — The Hacker News, Nov 13, 2026
- Java4Shell Flaw (CVE-2026-114938) Enables Pre-Auth RCE in Jira, Jenkins, Keycloak and More — The Hacker News, Nov 18, 2026
- 10-Year-Old NGINX HTTP/2 Flaw Lets Unauthenticated Attackers Execute Code on Servers — The Hacker News, Nov 18, 2026
- iPhone Users Targeted in Major Website Security Breach — MacRumors, Nov 17, 2026
- Optimizely hack: Everything we know so far — ITPro, Nov 17, 2026
- Hackers Breach N-able Cloud Server, Disrupting Email Security Management Across 3 Regions — The Hacker News, Nov 19, 2026
- Rapid7 Detected Exploitation of CVE-2026-90436 against Atlassian Confluence Data Center, N-able N-central also Exploited — Rapid7, Nov 16, 2026 (updated Nov 19)
- CISA Warns of 4 Exploited Flaws Linked to Ransomware, Cryptomining, and LLMjacking Attacks — The Hacker News, Nov 19, 2026
- Ubiquiti Warns of Critical UniFi Access Flaws Enabling Root Code Execution Attacks — Cyber Security News, Nov 19, 2026
- F5 Warns 6 Critical BIG-IP Flaws Can Crash Traffic in Specific Setups, May Allow RCE — The Hacker News, Nov 18, 2026
- Nvidia investigates ransomware claims as hackers threaten leak — American Bazaar, Nov 19, 2026