← cd ../blog

CVE-2026-76466: a static cluster secret gives root on Cisco Expressway

A hard-coded credential in Cisco Expressway's cluster database service (CVSS 9.8, CISA KEV) gives unauthenticated root on clustered nodes. Exposure, F6's in-the-wild findings, detection and the fix.

Cisco fixed CVE-2026-76466 on 21 October 2026 with a “not aware of any malicious use” note. A month later that has changed: Russian security firm F6 published an incident investigation on 18 November, and CISA added the bug to the KEV catalog today. It’s a credential problem in how clustered Expressway nodes trust each other, and it ends in root on the appliance that sits in your DMZ brokering voice and video. This is the second of two KEV additions I’m covering today, alongside the vCenter authentication bypass.

Scope and ethics. This post covers architecture, exposure checks, detection and remediation for systems you own or are authorized to assess. It doesn’t reproduce the static secret, the RPC calls, or any exploit steps, and it doesn’t link to the public proof of concept. Lab validation means confirming versions, cluster settings and firewall rules, not connecting to anyone’s cluster service.

At a glance

   
CVE CVE-2026-76466 (Cisco bug ID CSCww27732, advisory cisco-sa-vcs-static-cred-hr7EWZ8t)
Severity CVSS 3.1 9.8 Critical, AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H (Cisco PSIRT, via NVD)
CWE CWE-798 Use of Hard-coded Credentials
Affected Expressway-C and Expressway-E (virtual and appliance) on a vulnerable release, only when clustered
Fixed in X15.5.20. X14 and earlier: migrate to a fixed release
Exploited? Yes. CISA KEV, added 20 Nov 2026 (federal due date 23 Nov), forensic triage required. F6 documents a real intrusion. Cisco’s advisory (v1.0) still says it’s unaware of malicious use
Auth / interaction None / none
Workarounds None (Cisco)

Description: how the bug works

Cisco Expressway is the gateway pair that connects Cisco collaboration (Unified CM, Webex, video endpoints) to the outside world. Expressway-C sits on the internal network, Expressway-E sits in the DMZ, and together they provide Mobile and Remote Access (MRA) and business-to-business calling. For capacity and resilience, nodes of the same type are grouped into clusters that replicate configuration between peers.

That replication runs through what Cisco calls the database service. Cisco’s root-cause statement, via NVD, has two parts:

  1. “This vulnerability is due to the use of persistent cookies.” Read with the KEV wording (“an attacker with persistent cookies”), this means the database service accepts a cookie as proof that the caller is a cluster peer. Cisco classifies the bug as CWE-798 (hard-coded credentials), which indicates the cookie isn’t generated uniquely per deployment, so anyone who knows it can present it. Cisco doesn’t spell that out, and I’m not going to reproduce or describe the value.
  2. Insufficient client certificate validation. Cisco says that “when the TLS Verify Mode is set to Enabled for clustering,” client certificates are not validated. TLS on the cluster link therefore doesn’t stop a non-peer from connecting.

Combine them and an attacker who can reach the database service and has the cookie can connect, invoke “specific remote procedure call (RPC) commands,” and run commands as root on the underlying OS.

One piece of context from the field: F6’s indicators include an Erlang system log entry recording a connection to an Erlang node, and F6 recommends restricting TCP 4371–4372 to cluster peers. Erlang distributed nodes authenticate each other with a shared “cookie” and can execute code on each other by design. That fits Cisco’s wording closely, but it’s my inference. Cisco hasn’t named the technology.

Why the hard-coded framing matters. Hard-coded secrets don’t need to be cracked or phished. Once one has been published, every unpatched cluster with the port reachable is exposed, which matches what F6 saw: exploitation started after a public PoC appeared.

Impact

  • Root command execution on the Expressway node, with no credentials and no user interaction.
  • A pivot from the DMZ to the inside. In F6’s case, attackers compromised two internet-facing Expressway nodes, then used one of them to reach two more nodes of the same cluster on the internal network. Clustering is the trust relationship the bug abuses, so it’s also the lateral movement path.
  • Hands-on-keyboard follow-up. F6 says the attackers wrote Python tooling on the box for network scanning, SMB share discovery, SQL Server and WinRM access, and searching GitLab for credentials. Code comments suggested an AI assistant was used, and the scripts changed during the intrusion.

Which deployments are exposed

Configuration Status
Clustered Expressway-C or -E, X15.5.3 or earlier Vulnerable. Upgrade to X15.5.20
Clustered, X14 or earlier Vulnerable. Migrate to a fixed release
Standalone (non-clustered) Expressway Not affected by CVE-2026-76466. Still upgrade: the October hardening release (CVE-2026-76490 to -76497, up to CVSS 8.6) applies regardless of configuration and is also fixed in X15.5.20
X15.5.20 or later Fixed

Cisco’s table only names “X15.5.3 and earlier” and X15.5.20, so I can’t tell you anything about intermediate builds. If you run one, ask Cisco TAC.

Prerequisites: network reachability to the cluster database service, a vulnerable clustered node, and knowledge of the static secret. The KEV entry describes the attacker as having “persistent cookies and network access” to “a clustered device.”

Identifying affected systems

1. Version. In the web UI the software version appears on the status overview. On the CLI (SSH as admin):

xStatus SystemUnit

Look for the Version field in the Software block. Anything below X15.5.20 needs the upgrade.

2. Clustering. In the web UI, go to System > Clustering. If there are peer addresses, the node is clustered and in scope. Note the TLS verification mode setting as well. Cisco’s advisory ties the certificate weakness to it, but the fix is the upgrade either way, so don’t treat any value as a mitigation.

3. Reachability of the cluster ports. F6 names TCP 4371–4372. From a host that is not a cluster peer, on each network segment that matters to you (internet, DMZ, internal user ranges):

# Benign reachability check against your own Expressway nodes.
# Any "open" result from a non-peer is the exposure F6 saw exploited.
nmap -Pn -sT -p 4371,4372 expressway-e1.example.com expressway-c1.example.internal

Cisco’s Expressway IP port usage guide for your release lists the full set of intra-cluster ports; use it to build the allowlist.

Safe validation in an isolated lab

  1. In a lab, build a two-node cluster on your current production release. Run your scanner and confirm it flags CVE-2026-76466. Upgrade to X15.5.20 and confirm it no longer does.
  2. Apply a firewall rule that allows the cluster ports only between peers. Re-run the nmap check from a non-peer host and confirm the ports are filtered, then confirm replication still works (System > Clustering shows peers as active).
  3. Make a benign connection attempt to 4371 from a non-peer lab host and confirm your SIEM or firewall alert fires (see below).
  4. Practice your evidence collection: generate a diagnostic log bundle and a system snapshot, and confirm you know where they land and how to get them off the box.

Detection and threat hunting

Cisco hasn’t published IOCs. F6’s report lists two IP addresses, the Erlang log entry, and file path patterns for the attackers’ Python scripts; use F6’s report for the exact values. Behavioral leads:

  • Cluster-port connections from non-peers. Any session to TCP 4371–4372 from an address that isn’t a configured peer is suspicious. On Expressway-E that includes anything from the internet.
  • Erlang node connection log entries that don’t line up with known peers or maintenance windows (F6’s main host indicator).
  • Unexpected outbound traffic from Expressway. It talks SIP/H.323, media, Unified CM and Webex, not SMB (445), WinRM (5985/5986), SQL Server (1433) or your GitLab. Firewall logs showing those from an Expressway IP are a high-fidelity signal.
  • New scripts and processes. Python files on mounted disks, or Python processes running as root that aren’t part of the product.
  • Configuration drift. Admin accounts, peers or certificates that weren’t in your last change.

Collect evidence before you upgrade. CISA requires forensic triage for this entry, and F6 recommends compromise assessments even after patching, because attackers may have gotten in before the fix.

Remediation

  1. Upgrade every node in the cluster to X15.5.20. Cisco: “There are no workarounds that address this vulnerability.” X14 and earlier need a migration. Without a service contract, Cisco says to go through Cisco TAC for fixed software.
  2. Hunt first if the cluster ports were reachable from untrusted networks after the PoC became public, and preserve logs and snapshots.
  3. After a suspected compromise, rebuild and rotate. I’d treat a rooted node as untrustworthy: rebuild it from a fixed image, re-issue its certificates, and rotate any credentials it stores or brokers (directory bind accounts, Unified CM application users, admin passwords). Then hunt for the lateral movement F6 describes: SMB, WinRM, SQL and GitLab access from Expressway addresses.

Interim mitigation (if you can’t patch today)

  • Restrict TCP 4371–4372 to cluster peers only, as F6 recommends, on the host-side firewall and at the network firewall. This reduces exposure but isn’t a fix. Cisco lists no workarounds.
  • Make sure Expressway-E’s cluster ports aren’t reachable from the internet at all. Its public interface should expose only the collaboration ports you actually use.
  • Restrict egress from Expressway nodes to the destinations they need.

Edge appliances that broker trust between the internet and the inside keep turning up in KEV; see my posts on NetScaler, Ivanti Connect Secure and the 2026 RCE field guide.

References

Verify every version against the vendor advisory before acting. Published 20 Nov 2026.