Russian security firm F6 reports that attackers used CVE-2026-76466 to break into a Russian organization through two internet-facing Cisco Expressway appliances. It is the first exploitation of the flaw reported in the wild. The bug is rated CVSS 9.8 and gives an unauthenticated attacker root on clustered Expressway-C and Expressway-E nodes. A working proof-of-concept appeared on GitHub on November 5, and the intrusion F6 describes came soon after.
Cisco’s advisory (cisco-sa-vcs-static-cred-hr7EWZ8t) still says the vendor is not aware of malicious use. As of November 20 the CVE is not in the CISA KEV catalog. Treat both as lagging indicators: exploit code is public, the vulnerable service is reachable over the network, and successful exploitation gives you root.
What’s broken
Expressway clusters keep their configuration database in sync over Erlang distribution, with TLS on the cluster ports (TCP/4371–4372). Erlang distribution authenticates peers in two ways: a shared “magic cookie” and a node name that has to be on the allowed-node list. Any node that passes both checks can make RPC calls into the peer, including os:cmd-style execution. That makes it full remote code execution as the service’s user, which on Expressway is root.
Cisco says the cause is “persistent cookies” combined with insufficient client-certificate validation when clustering runs in TLS Verify Mode. Researcher write-ups and the public PoC give a more specific account, and CWE-798 (hard-coded credentials) is assigned to the CVE. According to those sources, the cm_server cluster-manager module adds a built-in maintenance node identity to the allowed-node set and installs a matching fixed cookie. Both values are compiled in, so they are the same on every install of an affected release.
The client-certificate side makes it worse. In TLS Permissive mode, which is the default, the cluster listener accepts a TLS client without validating its certificate. Anyone who can reach TCP/4372 and knows the built-in identity and cookie can join the distribution mesh as a trusted peer and execute commands as root. You don’t need credentials, a session, or user interaction.
Scope details:
- Affected: Expressway-C and Expressway-E, virtual and appliance, only when clustered. Standalone nodes are not in Cisco’s vulnerable set.
- Vulnerable releases: X15.5.3 and earlier. X14 and earlier are also affected and need a migration to a fixed train.
- Fixed: X15.5.20.
- Workarounds: Cisco says there are none.
The public PoC (dinosn/cve-2026-76466) was validated on Expressway-C X15.3.2 in a two-node cluster in TLS Permissive mode. It includes a --mode ping option that only checks whether a node accepts the maintenance identity, without running commands. Nobody has published tests of TLS Enforcing mode, Expressway-E, or X14. Until someone shows otherwise, assume all of them are exploitable.
The intrusion
F6 says the victim ran Expressway X15.5.3, one release below the fix, on two internet-exposed boxes. The attackers came in through the cluster database service. F6’s main host-level indicator is a nodeup entry in the Erlang system log (erlang.log) for a node that is not one of the cluster’s real peers. A nodeup event is the record of a new node joining the distribution mesh. Legitimate peers produce these too, so compare each nodeup against your actual cluster member list. Other reported indicators include the Erlang node name [email protected] and inbound connections to 4371/4372 from addresses outside the cluster. F6 has not publicly named the actor.
Expressway-E sits in the DMZ as the traversal edge for Webex, Jabber, and SIP/H.323 video. It holds TLS keys, connects to Unified CM and internal Expressway-C nodes, and often has paths into the internal voice network. A root foothold there puts the attacker past the perimeter.
What to do now
- Inventory. Find every Expressway-C/E node and record whether it is clustered and what release it runs. Clustered nodes on X15.5.3 or earlier are vulnerable.
- Upgrade to X15.5.20. If you’re on X14 or older, plan the migration now. No backport is coming.
- Filter TCP/4371–4372 immediately. Only cluster peers should be able to reach these ports. They should never be reachable from the internet, and they shouldn’t be reachable from general internal networks either. Alert on any connection from a non-peer address.
- Switch clustering to TLS Verify/Enforcing mode if your certificates support it. Cisco’s wording implies the cookie problem still exists there, but enforced mutual TLS adds a real hurdle on top of the static cookie.
- Hunt. Search
erlang.logon every node fornodeupevents from unexpected node names, especially[email protected]. Check for new local accounts, changed configuration, unexpected processes, and outbound connections from the appliance. If you were exposed and running a vulnerable release after November 5, assume compromise. Rebuild the node and rotate its TLS certificates and any credentials it stores for Unified CM, LDAP, and other integrations.
References
- Cisco advisory: cisco-sa-vcs-static-cred-hr7EWZ8t
- GitHub Advisory: GHSA-p2jr-qv63-hjch
- Public PoC and detector: dinosn/cve-2026-76466
- SecurityWeek: Hackers exploit Cisco Expressway vulnerability soon after PoC release
- Security Affairs: Hackers exploit Cisco Expressway flaw CVE-2026-76466 to breach Russian organization