Exploited Vulnerabilities

Cisco Expressway CVE-2026-76466 exploited: a hard-coded cluster cookie gave attackers root and a way inside

Cisco Expressway CVE-2026-76466 is exploited: a hard-coded cluster cookie gives root on clustered nodes. The attack chain, IOCs and how to check exposure.

By · · 6 min read

Cisco Expressway CVE-2026-76466 is now exploited in the wild: attackers used the critical flaw (CVSS 9.8) to get root on clustered Expressway X15.5.3 nodes at a Russian organization and then move into its internal network. Cisco fixed the bug on 21 October 2026. A public proof of concept appeared on GitHub on 5 November, the Russian security company F6 reported the first real intrusion on 18 November, and CISA added the CVE to its Known Exploited Vulnerabilities (KEV) catalog on 20 November.

What happened

Cisco’s advisory describes a vulnerability in the database service of Cisco Expressway Series that lets an unauthenticated, remote attacker with network access run operating system commands as root. Cisco says it is due to “the use of persistent cookies”, plus insufficient validation of client certificates when TLS Verify Mode is set to Enabled for clustering. The weakness is classed as CWE-798, use of hard-coded credentials.

Only clustered devices are affected: Expressway-C and Expressway-E, virtual and appliance. There is no workaround. Cisco credits Stefan Aleksic of Jane Street.

Expressway release First fixed release
X15.5.3 and earlier X15.5.20
X14 and earlier Migrate to a fixed release

This morning the advisory was still at version 1.0 and still said Cisco PSIRT was not aware of public announcements or malicious use. That wording is out of date. CISA’s KEV entry, added 20 November, gives federal agencies until 23 November and flags the entry for forensic triage, meaning agencies are expected to check for compromise, not just patch. The same KEV update added the VMware vCenter zero-day CVE-2026-123890.

F6 did not name the victim and did not attribute the attack to a known group.

The attack chain, step by step

Steps 1 to 3 come from the public proof of concept, steps 4 to 7 from F6’s investigation.

  1. Reach the cluster port. Expressway uses Erlang/OTP for its cluster database. According to the PoC, each clustered node runs an Erlang distribution listener on TCP 4372 behind TLS, and that service runs as root.
  2. Present the built-in identity. The PoC says the cluster manager module, cm_server, admits a hard-coded maintenance node named wombat@127.0.0.1 (a legacy TANDBERG prefix) and installs a matching hard-coded cookie, wombat. Both are identical on every installation of an affected release. In TLS Permissive mode, which the PoC author calls the factory default, client certificates are not validated.
  3. Run commands as root. After the Erlang handshake, an rpc:call to os:cmd executes any shell command. The author tested it on Expressway-C X15.3.2 in a two-node cluster and got id -u = 0. Expressway-E, X14 and TLS Enforcing mode were not tested.
  4. Edge first. Shortly after the PoC went public, the attackers compromised two internet-facing Expressway X15.5.3 devices.
  5. Pivot over the cluster link. Through one of those devices they reached two more cluster nodes on the internal network. Four nodes in total.
  6. Tooling on the box. With root, they staged custom Python scripts to scan the network, find accessible SMB shares, talk to SQL Server and WinRM, and search GitLab repositories for credentials and other sensitive data to exfiltrate.
  7. AI-assisted edits mid-intrusion. F6 says the attackers modified the scripts during the attack, and the code had comments and fix notes indicating the use of an AI assistant.

Why it worked

A shared secret that is not secret. An Erlang cookie is normally the only authentication on a distribution port. If it is compiled into the product, anyone with the PoC or a copy of the firmware knows it for every deployment. TLS encrypts the session, but without client certificate validation it does not prove who is connecting.

Cluster design does the lateral movement for you. Expressway-E usually sits at the edge so remote users can register, and the cluster nodes have to talk to each other. A node that falls at the perimeter is already allowed to speak to its internal peers on exactly the vulnerable port.

Two weeks from patch to PoC. The fix shipped 21 October, the PoC 5 November, the intrusion followed. It is the same pattern as the Check Point zero-day chain: an edge appliance with root access becomes the jump host.

What defenders should do

  • Upgrade every node to X15.5.20. A cluster with one unpatched node still has an open door. X14 and earlier need a migration plan, not a waiting list.
  • Restrict TCP 4371 and 4372 to the IP addresses of trusted cluster peers on network and host firewalls. F6 recommends this if you cannot patch immediately. It reduces exposure; Cisco lists no workaround.
  • Treat any exposed vulnerable node as possibly compromised. The fix removes the identity, not an attacker who already has root. Check before and after the upgrade.
  • Watch the inside. Alert on SMB, WinRM, SQL Server and GitLab connections that originate from Expressway nodes.

How to check if you are affected

  1. Version and cluster: check each node’s release. Clustered and X15.5.3 or earlier means vulnerable. Standalone nodes are outside Cisco’s affected list.
  2. Exposure: from a host that is not a cluster peer, test whether TCP 4371 or 4372 answers. Any success from a user VLAN or the internet is the first thing to close.
  3. Erlang log: F6’s key indicator is the entry nodeup,'wombat@127.0.0.1' in erlang.log on the Expressway. If you find it, check other logs and artifacts to confirm compromise.
  4. Files: Python scripts under paths matching /mnt/harddisk/i{0-9}{4}[a-zA-Z]?/*.py (four digits, optional letter).
  5. Network IOCs (defanged): 66.179.31[.]32 and 18.130.47[.]116.
  6. Downstream: review GitLab access logs, SQL Server logins and WinRM sessions from Expressway IP addresses since early November, and rotate any credentials stored in repositories those nodes could reach.

My take

Hard-coded credentials are old news. The interesting part is the timeline. Cisco disclosed on 21 October with a short, vague advisory, and the details that made exploitation trivial arrived two weeks later in a public repository. If your patch process ranks a voice and video gateway below a VPN concentrator, this case is the argument to change it.

The second lesson is segmentation inside the cluster itself. Cluster peers need to reach each other, but nothing else needs port 4372. A one-line firewall rule would have turned this from “attackers inside the network” into “one compromised edge box”. More cases like this are in the exploited vulnerabilities category.

Questions worth asking this week: do you know which of your Expressway nodes are clustered, who can reach 4371 and 4372, and would anyone notice if a collaboration appliance started querying your GitLab?

Sources

  1. Cisco Expressway Series Remote Code Execution Vulnerability (cisco-sa-vcs-static-cred-hr7EWZ8t) · Cisco, 21 October 2026
  2. F6 records the first case of exploitation of the critical CVE-2026-76466 vulnerability in Cisco Expressway (in Russian) · F6, 18 November 2026
  3. CISA Adds Four Known Exploited Vulnerabilities to Catalog · CISA, 20 November 2026
  4. CVE-2026-76466: Cisco Expressway cluster database root RCE (public proof of concept) · GitHub, dinosn, 5 November 2026
  5. Hackers Exploit Cisco Expressway Vulnerability Soon After PoC Release · SecurityWeek, 19 November 2026
  6. Hackers Exploit Critical Cisco Expressway Flaw, Breach Four Devices and Reach Internal Network · The Hacker News, 20 November 2026
  • cisco
  • cisco-expressway
  • cve-2026-76466
  • erlang
  • cwe-798
  • f6
  • unified-communications