This article was first published on the Rokkhe blog on Hashnode. It was drafted with AI assistance and reviewed and edited by the Rokkhe team; all facts were checked against the linked public sources as of 21 Nov 2026.
What attackers actually did between 12 and 21 November 2026, and what to check on your side before Monday. Built from vendor advisories, CISA's KEV catalog and the incident write-ups, as of 21 Nov 2026. By the Rokkhe research team.
In four of the six incidents below, attackers did not need a new trick. They came in through access someone had granted on purpose: a vendor's AWS role, an old SOAP API, a release pipeline, an AI gateway key.
The other two were straight exploitation of management systems, one of them a zero-day that was used for more than two months before a patch existed. Below is what happened, who is exposed, how to check, and what to do now, with a copy-paste checklist at the end. Every claim links to where it came from.
TL;DR
- VMware vCenter (CVE-2026-123890). An authentication bypass that gives an attacker
administrator@vsphere.local. Exploited as a zero-day since early September, patched 19 Nov, in CISA KEV since 20 Nov. Patch vCenter now and wall off its management ports. - Apache Axis 1.x (CVE-2026-90436) in Confluence Data Center and N-able N-central. Unauthenticated code execution through an old SOAP API that is off by default. Rapid7 saw an attacker go from the first request to a cloud data warehouse in about four hours. Upgrade or turn the Remote API off.
- Cisco Expressway (CVE-2026-76466). Root on clustered Expressway devices. F6 investigated a real intrusion that followed a public proof of concept. Upgrade to X15.5.20.
- LiteLLM (CVE-2026-40217). In CISA KEV since 18 Nov after Google reported attackers turning ordinary LiteLLM virtual keys into admin keys. Upgrade to 1.84.0 or later, not just 1.83.10.
- Qubole cross-account AWS roles, and the Chase, Visa and Mastercard notices. An attacker used a vendor's AWS identity to assume roles its customers had set up, many of them unused. This week three payment and banking names confirmed third-party incidents. Card numbers were not reported affected. Find and remove vendor roles you no longer use.
- npm packages shipped from real release pipelines.
@wppconnect-team/wppconnect2.4.1 and@wppconnect/server2.10.82 sent secrets to Telegram.@leadstack/crm-connectinstalled an AI agent as a backdoor. Provenance told you where a package was built, not that anyone reviewed it.
This week in attacks, at a glance
| # | Incident | Exploited? | Who should look | First step |
|---|---|---|---|---|
| 1 | VMware vCenter auth bypass, CVE-2026-123890 | Yes, zero-day since early Sep (GTIG); CISA KEV 20 Nov | Anyone running vCenter 7.0, 8.0, VCF or vSphere Foundation | Patch per VMSA-2026-0009; restrict TCP 2012, 443, 5480 to admin hosts |
| 2 | Apache Axis 1.x RCE, CVE-2026-90436 (Confluence DC, N-central) | Yes, from 13 Nov (Rapid7); CISA KEV 18 Nov | Confluence DC 1.x to 9.x with Remote API on; N-central with Legacy API on; any Java app bundling Axis 1.x | Confluence 9.2.28 or 10.x; N-central 2026.4 HF3 |
| 3 | Cisco Expressway root RCE, CVE-2026-76466 | Yes, one investigated case (F6); CISA KEV 20 Nov | Clustered Expressway-C and Expressway-E | Upgrade to X15.5.20; limit TCP 4371 to 4372 to cluster peers |
| 4 | LiteLLM sandbox escape, CVE-2026-40217 | CISA KEV 18 Nov, after GTIG reporting | Self-hosted LiteLLM proxy | Upgrade to 1.84.0 or later; rotate virtual keys |
| 5 | Qubole cross-account AWS role abuse; Chase, Visa, Mastercard third-party notices | Yes, 14 to 26 Oct (Wiz); bank and card notices 19 to 20 Nov | Any AWS account with a role that trusts a vendor account | Remove unused vendor roles; add source IP conditions |
| 6 | npm packages from trusted pipelines: wppconnect, @leadstack/crm-connect | Malicious versions published 15 to 16 Nov (StepSecurity, SafeDep) | Node.js projects and Docker hosts using these packages | Pin clean versions; rotate secrets on any host that loaded them |
CISA added 11 vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog between 12 and 20 November. Catalog version 2026.11.20 lists 1,790 entries (CISA KEV, KEV data). The rest of this week's additions are listed further down.
1. VMware vCenter: a zero-day that hands out the admin account (CVE-2026-123890)
What happened
On 19 November Broadcom published VMSA-2026-0009, which fixes CVE-2026-123890, an authentication bypass in the VMware Directory Service inside vCenter, rated 9.8 under CVSS 3.1 (Broadcom Q&A, NHS England). The same day, Google Threat Intelligence Group (GTIG) and Mandiant published how a suspected China-nexus group they track as ARMORY CASTLE (formerly UNC5221) used it as a zero-day in early September 2026 (GTIG).
GTIG describes the bug as an unauthenticated SRP-6a session key disclosure over DCE/RPC. Exploiting it creates a session as administrator@vsphere.local. In the intrusions Mandiant investigated, the group first broke into a BeyondTrust Remote Support appliance with a different zero-day (CVE-2026-102764), moved to an internal Linux appliance, scanned for VMware ports including TCP 2012, then used a Go tool called BONESAW to create a vCenter SSO administrator account and copy the directory database (GTIG, BleepingComputer). From there they forged SAML tokens, reached ESXi hosts and pulled files such as Active Directory databases out of virtual machines (BleepingComputer).
CISA added the CVE to KEV on 20 November with a 23 November federal deadline (CISA). Broadcom's advisory does not itself mention exploitation in the wild, and Mandiant's CTO said the company investigated multiple incidents (SecurityWeek).
Who is exposed
Every vCenter below the fixed builds: vCenter 8.0 before 8.0 U3m, vCenter 7.0, VMware Cloud Foundation and vSphere Foundation 9.1.x before 9.1.1.0200 and 9.0.x before 9.0.2.0200, VCF 5.x before 5.2.5, and the Telco Cloud vCenter components (Cyber Centre, NHS England). vCenter 7.0 is past end of support, so NHS England points 7.0 users to Broadcom if they have an extended support contract (NHS England). Enhanced Linked Mode does not change the answer (BleepingComputer).
vCenter is rarely on the internet. That did not help here: the attacker reached it from inside, from an appliance that was already compromised.
How to check
- Compare your vCenter version and build (vSphere Client, Help, About) with the fixed list in VMSA-2026-0009.
- From an ordinary server or user subnet, test whether you can reach vCenter on TCP 2012, 443 and 5480. If you can, so could the attacker's foothold.
- Look for SSO administrator accounts you did not create, SSH being turned on through the appliance management API, and outbound connections from the vCenter appliance. GTIG lists these behaviours and publishes YARA rules for the tools (GTIG).
What to do now
- Patch. Broadcom treats this as an emergency change and says there is no official workaround (Broadcom Q&A, BleepingComputer).
- Restrict vCenter management ports with a network firewall, not only the appliance's own firewall. GTIG explains why: the built-in firewall is managed through the same management interface an attacker can abuse, and it does not log denied connections (GTIG).
- If you also run BeyondTrust Remote Support, check it is on the BT26-05 fix. That appliance was the way in (GTIG).
- If you find signs of compromise, treat SSO signing keys as stolen. Patching does not undo a forged token.
2. Apache Axis 1.x: an old SOAP library inside Confluence and N-central (CVE-2026-90436)
What happened
On 6 November Apache disclosed CVE-2026-90436 in Axis 1.x, a Java SOAP library that reached end of life years ago. A crafted SOAP fault makes Axis load and instantiate a class named by the attacker before any authentication handler runs. With a suitable class on the classpath, that is remote code execution. Apache will not ship a fixed 1.x release (oss-security).
It then turned out that Atlassian Confluence Data Center and N-able N-central bundle Axis 1.x. Both need a non-default setting to be exposed (Rapid7). A proof of concept against N-central was published on 12 November. On 13 November Rapid7's MDR team responded to exploitation against a Confluence server, and N-able reported exploitation against N-central the same day. A Metasploit module followed on 17 November and CISA added the CVE to KEV on 18 November (Rapid7).
Rapid7's account of the Confluence intrusion is worth reading in full. Unauthenticated POST requests to /plugins/servlet/soap-axis1/confluenceservice-v2 returned HTTP 500, yet the Confluence Java process started Bash and curl within seconds. Just over three minutes later the attacker was logged in to MySQL with Confluence's own database credentials, searching pages and attachments for credentials, tokens, VPN configs and private keys. About four hours in, they had a service account credential and were querying a production cloud data warehouse. About twelve hours in, they signed in to Microsoft 365 as an integration service account using legacy authentication. Rapid7 says the structure of the commands suggests AI-assisted automated tooling (Rapid7).
Who is exposed
- Confluence Data Center 1.x to 9.x with "Remote API (XML-RPC & SOAP)" turned on. It has been off by default for new installs since Confluence 3.0, was removed in 10.0 (Atlassian). Rapid7 notes Atlassian Cloud is not affected (Rapid7).
- N-able N-central on-premises servers with "Legacy API" turned on. N-able says it is off by default (N-able, Rapid7).
- Your own Java apps that still ship
axis-1.4.jaror older and expose a SOAP endpoint.
Previdian reported 50 exploitation attempts from 14 IP addresses between 12 and 19 November, without being able to say whether they succeeded (SecurityWeek).
How to check
- Confluence: browse to
<base_url>/plugins/servlet/soap-axis1/confluenceservice-v2. Atlassian says a correctly mitigated instance answers "HTTP Status 403 - Forbidden" (Atlassian). - Search Confluence access logs for POST requests to that path, especially ones that returned 500, and EDR data for
bashorcurlstarted by the Confluence Java process (Rapid7). - Search your own build artefacts for Axis 1.x jars.
What to do now
- Confluence: upgrade to 9.2.28 or a 10.x release. If you cannot today, clear "Remote API (XML-RPC & SOAP)" under General Configuration, Further Configuration. It takes effect without a restart, but scripts and add-ons that use the API may stop working (Atlassian).
- N-central: upgrade to 2026.4 HF3 (2026.4.1.39), released 16 November, which removes the Legacy API and fixes seventeen CVEs, and follow N-able's Agent and Probe upgrade guidance (Rapid7, N-able).
- If the endpoint was reachable before you patched, assume the credentials Confluence holds were read: its database password, application links, and anything people pasted into pages and attachments. Rotate them.
3. Cisco Expressway: root on clustered devices (CVE-2026-76466)
What happened
Cisco disclosed CVE-2026-76466 on 21 October, rated 9.8. It sits in the database service of clustered Expressway devices: persistent cookies plus missing client certificate validation let an attacker with network access call RPC commands and run commands as root (Cisco).
On 18 November the Russian security company F6 described the first exploitation it had investigated, against a Russian organisation, shortly after a public proof of concept appeared. The attackers took two internet-facing Expressway X15.5.3 devices, used one to reach two more cluster nodes on the internal network, then ran Python scripts to scan the network, look for SMB shares, talk to SQL Server and WinRM, and search GitLab for credentials. F6 says the scripts were edited during the attack and contained comments that point to an AI assistant (F6, SecurityWeek). CISA added the CVE to KEV on 20 November (CISA).
Cisco's advisory, at the time SecurityWeek checked, still said Cisco was not aware of malicious use (SecurityWeek).
Who is exposed
Expressway-C and Expressway-E, virtual or appliance, configured in a cluster and running X15.5.3 or earlier. X14 and earlier need to move to a fixed release (Cisco).
How to check
- Version in the Expressway web interface.
- F6's main sign of exploitation: the line
nodeup,'wombat@127.0.0.1'in the Erlang system log (erlang.log), plus Python files under paths matching/mnt/harddisk/i<4 digits>/(F6).
What to do now
- Upgrade to X15.5.20. Cisco lists no workaround (Cisco).
- Until then, F6 suggests allowing TCP 4371 to 4372 only from the IP addresses of your own cluster nodes (F6).
4. LiteLLM: when an AI gateway key becomes an admin key (CVE-2026-40217)
What happened
LiteLLM is an open source proxy that sits between your apps and model providers and hands out "virtual keys". CVE-2026-40217 is a sandbox escape in POST /guardrails/test_custom_code: code submitted there runs in the proxy process, which is root in the default Docker image. Reaching the endpoint needs a proxy-admin key by default. It was fixed in April in v1.83.10-stable (GitHub advisory, NVD).
CISA added it to KEV on 18 November, with a 2 December deadline (NVD). The exploitation evidence traces back to GTIG's 2 November report on "LLMjacking", where a follow-on group turned ordinary virtual keys into PROXY_ADMIN credentials on a victim's LiteLLM gateway (GTIG).
Where sources disagree
GTIG's report describes the bug it saw as a user ID rebind in /key/regenerate and labels it CVE-2026-40217. Corgea points out that these are two different bugs: the sandbox escape (the actual CVE) and the /key/regenerate rebind, which had no CVE of its own and was fixed in 1.84.0 in May. Chained, they go from a leaked inference key to root on the gateway (Corgea). We have not seen a statement from BerriAI or GTIG settling it, so the safe reading is to fix both.
Who is exposed
Self-hosted LiteLLM proxies from 1.81.8 up to 1.83.10 for the CVE, and any build before 1.84.0 for the rebind (GitHub advisory, Corgea). If you use LiteLLM only as a Python library inside your app, with no proxy server, these endpoints are not exposed.
What to do now
- Upgrade the proxy to 1.84.0 or later. Pin the container image by digest, since Corgea reports the
v1.83.10-stableimage was rebuilt after the patch (Corgea). - If you cannot upgrade today, block
POST /guardrails/test_custom_codeat your reverse proxy, as the advisory suggests (GitHub advisory). - List every key with admin rights and check when it was created. Rotate the upstream provider keys the gateway holds, and look at your model spend for usage you do not recognise. LLMjacking is about running up someone else's bill.
5. Qubole's AWS roles, and the Chase, Visa and Mastercard notices
What happened
Between 14 and 26 October, an attacker used IAM credentials from an AWS account that Qubole's platform uses to assume roles in its customers' AWS accounts, according to Wiz, which investigated with Sentra and Mandiant (Wiz). These were normal cross-account roles that customers had created during onboarding, with an External ID check. The attacker presented the right External ID every time, then listed resources, read Lambda environment variables, Secrets Manager and SSM Parameter Store, tested the credentials it found with TruffleHog, and downloaded S3 data (Wiz).
The detail that matters most: in most of the environments Wiz looked at, the roles were not in use. Sentra says they had been dormant since spring and were still able to reach sensitive data (Wiz, Sentra). Sentra is clear that the companies hit had set the integrations up the way AWS recommends (Sentra).
A group calling itself n0n claimed the intrusion (Wiz). This week the story reached household names. On 19 November JPMorgan Chase said third parties that provide services to it were attacked and that marketing data, which may include some customer names and email addresses, was accessed. Visa said a third-party provider in Asia Pacific was involved and no card numbers were affected. Mastercard said a provider exposed a limited number of email addresses from a small subset of cardholders in one benefit programme (The Register, AP). None of the three named its vendors. BleepingComputer reports that Qubole is suspected to be the provider involved, based on n0n's claims; that link is not confirmed by the companies (BleepingComputer).
Who is exposed
Directly, AWS accounts that ever integrated with Qubole. More broadly, any AWS account with a role that trusts another company's AWS account: data platforms, monitoring, backup, cost tools. Each one is a standing key held by someone else.
How to check
- Use IAM Access Analyzer's external access findings, or the CLI commands in the checklist below, to list every role that trusts an outside account.
- For Qubole specifically, Wiz suggests reviewing CloudTrail from 1 October for
AssumeRolecalls from account805246085872that do not come from AWS addresses, and for the IAM userAIDA3W7DIFLYOLTK63SFS(Wiz). - Check whether S3 data events are logged for sensitive buckets. Wiz notes that CloudTrail management events show the role assumption but not the object reads (Wiz).
What to do now
- Delete integration roles you no longer use, or remove the vendor from their trust policy.
- Where the vendor supports it, trust a specific vendor principal instead of the whole account, and add an
aws:SourceIporaws:SourceVpcecondition. Wiz says a source IP condition alone would have blocked every role assumption it observed (Wiz). - Treat the External ID as non-secret. It appears in plain text in CloudTrail (Wiz).
- If a role was assumed by the attacker, rotate every secret it could read and revoke its active sessions. Rotating a credential does not end sessions already issued (Wiz).
- If you are a Chase, Visa or Mastercard customer: the exposed data reported so far is names, emails and some phone numbers, which is the kind of data used for convincing phishing. Treat unexpected messages about your card or account with care and go to the bank's app or website yourself.
6. npm packages published by the projects' own pipelines
What happened
On 15 and 16 November, @wppconnect-team/wppconnect 2.4.1 and @wppconnect/server 2.10.82 went out through the projects' own GitHub Actions trusted-publishing workflows. The library release carried a valid SLSA provenance attestation. Both included a "diagnostics" module that uploads environment variables, .env and config files, and WhatsApp session tokens to a Telegram bot. Infected wppconnect/wppconnect-server Docker images were built from the same commits (StepSecurity, Corgea). The code runs on import or server start, not at install, so npm install --ignore-scripts does not stop it (StepSecurity).
The commits were made through a project member account. StepSecurity and others say they have not established how that access was obtained, and the evidence does not show who inserted the code (StepSecurity, TSN). The maintainers rolled latest back to 2.4.0 and 2.10.81 and deprecated the bad versions within hours of the public report (GitHub issue #2968, GitHub issue #2734).
The same day, a package called @leadstack/crm-connect, posing as a CRM client, installed Hermes Agent, an open source AI agent that can run shell commands, connected it to the attacker's Telegram bot and registered it as a systemd service. When npm install ran as root, the agent ran as root. SafeDep flagged it within minutes of the last release and npm removed it (SafeDep).
Who is exposed
- Anyone who installed or ran
@wppconnect-team/wppconnect2.4.1 or@wppconnect/server2.10.82, or pulled thewppconnect/wppconnect-serverimages named in the advisories on 16 November (Corgea). - Anyone who installed any of the
@leadstack/crm-connectversions listed by SafeDep (SafeDep).
What to do now
- Pin
@wppconnect-team/wppconnectto 2.4.0 and@wppconnect/serverto 2.10.81, and pull Docker images by digest (Corgea). - On any machine that loaded a bad version, treat every environment variable and config file in the working directory as exposed, rotate those secrets, and log out the WhatsApp sessions (GitHub issue #2734).
- For
@leadstack/crm-connect, look for the systemd unitsleadstack-routing.serviceandleadstack-ingest-worker.serviceand the~/.leadstack/directories SafeDep lists, and rebuild the host if you find them (SafeDep). - Add a cooldown before new versions of dependencies reach production. Both bad wppconnect versions and the leadstack package were flagged within hours of release.
Also this week
- N-able Mail Assure and SpamExperts Hosted Cloud. A service disruption that began on 16 November was traced to unauthorised access to one server. Mail flow kept working, but the portal, API, quarantine and logs were taken offline. N-able says it found no evidence customer data or mailboxes were accessed, and that it revoked customer API tokens, invalidated OAuth tokens and reset local user passwords. If you integrate with it, expect to reissue tokens (The Hacker News, SecurityWeek).
- Apache Roller (CVE-2026-129448). Unauthenticated code execution through XML-RPC in Roller 4.0.0 to 6.1.4. Apache says it is actively exploited, and disabling XML-RPC in the admin interface does not prevent it. Upgrade to 6.1.6 or later (SecurityWeek, Apache Roller).
- SAP NetWeaver Enterprise Portal (CVE-2025-42884). An unauthenticated JNDI injection fixed by SAP Security Note 3660969 on 11 November 2025, added to KEV on 20 November 2026 (CISA, SecurityBridge). If that note is not applied, this is a year-old gap.
- Other KEV additions 12 to 20 November: Check Point (CVE-2026-123039), UnrealIRCd (CVE-2026-114713), Perforce Puppet Enterprise (CVE-2026-115731), Synology DSM (CVE-2024-10441) and REDCap (CVE-2026-90817) (CISA KEV).
- nginx2root. Not exploited as far as anyone has confirmed, but if you run nginx with HTTP/2, our nginx2root guide covers the fix.
Patterns this week
Trust you granted is the attack surface. A vendor's AWS role, a legacy SOAP API someone enabled years ago, a release workflow, an AI gateway key. None of these is a bug in your code, and none shows up in a scan of your public website. Every one shows up in an inventory, if you have one.
Proof of concept to intrusion is now days, not months. Axis: proof of concept for N-central on 12 November, exploitation seen on 13 November (Rapid7). Expressway: a confirmed intrusion shortly after a public exploit (F6). A monthly patch cycle cannot keep up with that for anything reachable.
Attackers are using AI tooling, and it shows in the logs. Rapid7 saw command structure that suggests AI-assisted automation. F6 found AI assistant comments in the attackers' scripts. One npm package installed an AI agent as its backdoor (Rapid7, F6, SafeDep). The practical effect is speed: three minutes from code execution to the database in the Confluence case.
Management planes are the prize. vCenter, N-central, an email security cloud, an AI gateway. Each one holds keys to many other systems, so one foothold there is worth a lot.
A signed build is not a reviewed build. The wppconnect release had valid provenance. Provenance tells you where a package was built. It does not tell you a second person read the change.
Check if you're affected: copy-paste checklist
Run these against systems you own. Replace the example hostnames.
# 1. vCenter (CVE-2026-123890): from a normal server or user subnet, # these should FAIL. If they connect, restrict them. nc -zv -w 3 vcenter.example.internal 2012 nc -zv -w 3 vcenter.example.internal 5480 # Then compare your build (vSphere Client > Help > About) with VMSA-2026-0009. # 2. Confluence (CVE-2026-90436): expect 403 once Remote API is off or patched curl -s -o /dev/null -w '%{http_code}\n' \ https://confluence.example.com/plugins/servlet/soap-axis1/confluenceservice-v2 # Look for past exploit attempts in the access logs grep 'soap-axis1/confluenceservice-v2' /opt/atlassian/confluence/logs/conf_access_log* # 3. Axis 1.x bundled in your own Java apps find / -name 'axis-1*.jar' -o -name 'axis.jar' 2>/dev/null # 4. Cisco Expressway (CVE-2026-76466): on each clustered node, F6's indicator grep -r "nodeup,'wombat@127.0.0.1'" /path/to/erlang.log # 5. LiteLLM proxy (CVE-2026-40217 + key rebind): want 1.84.0 or later pip show litellm | grep -i '^version' docker ps --format '{{.Image}}' | grep -i litellm # 6. AWS: roles that trust another AWS account aws accessanalyzer create-analyzer --analyzer-name external-access --type ACCOUNT aws accessanalyzer list-findings --analyzer-arn <arn-from-above> \ --filter '{"resourceType":{"eq":["AWS::IAM::Role"]}}' # Qubole specifically: AssumeRole calls from its account since 1 Oct aws cloudtrail lookup-events --start-time 2026-10-01 \ --lookup-attributes AttributeKey=EventName,AttributeValue=AssumeRole \ | jq -r '.Events[].CloudTrailEvent | fromjson | select(.userIdentity.accountId=="805246085872") | [.eventTime, .sourceIPAddress, .userIdentity.principalId] | @tsv' # 7. npm: bad versions in any lockfile grep -rE 'wppconnect-2\.4\.1\.tgz|server-2\.10\.82\.tgz|@leadstack/crm-connect' \ --include=package-lock.json --include=yarn.lock --include=pnpm-lock.yaml . # and the leadstack backdoor on hosts systemctl list-units --all | grep -i leadstack ls -d ~/.leadstack ~/.config/leadstack 2>/dev/null Notes: cloudtrail lookup-events covers 90 days of management events in one region per call, so run it in each region where you have integration roles. The Expressway log path differs by version, so search for erlang.log first.
The short list for this weekend
- [ ] vCenter patched to the VMSA-2026-0009 fixed build, ports 2012, 443 and 5480 reachable only from admin hosts
- [ ] Confluence on 9.2.28 or 10.x, or Remote API off, and the 403 test passes
- [ ] N-central on 2026.4 HF3, Agents and Probes upgraded
- [ ] Expressway clusters on X15.5.20, TCP 4371 to 4372 limited to cluster peers
- [ ] LiteLLM proxy on 1.84.0 or later, admin keys reviewed, provider keys rotated if in doubt
- [ ] Every AWS role trusting an outside account reviewed; unused ones deleted
- [ ] Lockfiles checked for the bad npm versions; secrets rotated where they ran
- [ ] Apache Roller on 6.1.6 or later; SAP Note 3660969 applied
- [ ] Any system that was exposed before patching checked for compromise, not just patched
FAQ
What were the biggest cyberattacks this week (November 2026)?
The VMware vCenter zero-day used by ARMORY CASTLE (CVE-2026-123890), exploitation of Apache Axis in Confluence Data Center and N-central (CVE-2026-90436), a Cisco Expressway intrusion (CVE-2026-76466), LiteLLM in CISA KEV (CVE-2026-40217), abuse of Qubole's cross-account AWS roles with follow-on notices from Chase, Visa and Mastercard, and npm packages that shipped secret stealers through their own release pipelines.
Is CVE-2026-123890 in VMware vCenter being exploited?
Yes. GTIG and Mandiant report ARMORY CASTLE used it in early September 2026, before a patch existed, and CISA added it to KEV on 20 November 2026. Broadcom's fix is in VMSA-2026-0009 and there is no official workaround.
How do I know if my Confluence is vulnerable to CVE-2026-90436?
You are exposed if you run Confluence Data Center 9.x or older with "Remote API (XML-RPC & SOAP)" turned on. Browse to /plugins/servlet/soap-axis1/confluenceservice-v2. If you get a 403, the API is off. Confluence 10.x and Atlassian Cloud are not affected.
Were card numbers stolen in the Chase, Visa and Mastercard incidents?
Not according to the companies. Chase described marketing data that may include names and email addresses, Visa said no card numbers were affected, and Mastercard described a limited number of email addresses. Expect phishing that uses those details.
Does npm provenance mean a package is safe?
No. Provenance proves which workflow built the package. The wppconnect 2.4.1 release had valid provenance and still contained a secret stealer, because the change that added it was never reviewed.
What is the fastest useful thing to do today?
Find out whether you run vCenter, Confluence with the Remote API on, clustered Expressway, or a LiteLLM proxy, and patch those first. Then list the AWS roles that trust outside accounts and delete the ones nobody uses.
Where Rokkhe helps
We did not find any of these bugs. Mandiant, GTIG, Rapid7, F6, Wiz, Sentra, StepSecurity, SafeDep, Corgea and the vendors' own teams did that work, and the incident write-ups above are generous with detail. Thanks to all of them.
What we do is the part this week kept pointing at: knowing what you actually expose. Rokkhe runs AI-driven penetration testing on the systems you own and authorise us to test, finds the forgotten doors (old APIs left on, management ports reachable from the wrong subnet, vendor roles nobody uses), and our SOC watches for the signals above while you patch. Offense and defense in one loop, so a finding becomes a detection the same week.
If you want a second pair of eyes on your exposure this week, talk to us at rokkhe.com.
Sources
Vendor and project advisories
- Broadcom VMSA-2026-0009 Q&A: https://github.com/vmware/vcf-security-and-compliance-guidelines/blob/main/security-advisories/vmsa-2026-0009/README.md
- Atlassian, CVE-2026-90436 in Confluence Data Center: https://confluence.atlassian.com/security/cve-2026-90436-remote-code-execution-in-confluence-data-center-1874560038.html
- Apache Axis disclosure on oss-security: https://www.openwall.com/lists/oss-security/2026/11/06/12
- N-able vulnerability remediation update: https://www.n-able.com/blog/n-able-vulnerability-remediation-update
- Cisco advisory cisco-sa-vcs-static-cred-hr7EWZ8t: https://www.cisco.com/c/en/us/support/docs/csa/cisco-sa-vcs-static-cred-hr7EWZ8t.html
- LiteLLM GitHub advisory GHSA-wxxx-gvqv-xp7p: https://github.com/advisories/GHSA-wxxx-gvqv-xp7p
- Apache Roller security: https://roller.apache.org/security.html
- wppconnect issue #2968: https://github.com/wppconnect-team/wppconnect/issues/2968
- wppconnect-server issue #2734: https://github.com/wppconnect-team/wppconnect-server/issues/2734
Government and CERT
- CISA KEV catalog: https://www.cisa.gov/known-exploited-vulnerabilities-catalog
- CISA KEV data: https://github.com/cisagov/kev-data
- CISA alert, 20 Nov 2026: https://www.cisa.gov/news-events/alerts/2026/11/20/cisa-adds-four-known-exploited-vulnerabilities-catalog
- CISA alert, 18 Nov 2026: https://www.cisa.gov/news-events/alerts/2026/11/18/cisa-adds-four-known-exploited-vulnerabilities-catalog
- NVD, CVE-2026-40217: https://nvd.nist.gov/vuln/detail/cve-2026-40217
- Canadian Centre for Cyber Security AV26-1185: https://www.cyber.gc.ca/en/alerts-advisories/vmware-security-advisory-av26-1185
- NHS England CC-4899: https://digital.nhs.uk/cyber-alerts/2026/cc-4899
Incident research
- GTIG and Mandiant, ARMORY CASTLE exploits vSphere: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-vsphere
- GTIG, LLMjacking: https://cloud.google.com/blog/topics/threat-intelligence/llmjacking-monetizing-compromised-cloud-models
- Rapid7, CVE-2026-90436 explo...

