By ByteBot
Cisco disclosed CVE-2026-20212 on September 2 — a CVSS 9.8 unauthenticated remote code execution flaw in Nexus 9000 switches built on Silicon One ASICs. No credentials. No existing foothold. An attacker with network access to TCP port 43210 or 43211 gets root on the switch routing your GPU cluster traffic. Patches are available. Two interim mitigations exist. “No known active exploitation” is not a reason to wait — it is a starting gun.
What the Flaw Actually Does
The vulnerability, classified as CWE-1327 (Binding to an Unrestricted IP Address), exists because TCP ports 43210 and 43211 are exposed through the default Layer 3 VRF on affected switches. Any host with network connectivity to a locally configured address on those ports can connect directly to an internal Silicon One hardware service. Send specially crafted input and it executes as code with root-level OS privileges.
There is a secondary failure mode: successful exploitation can crash the S1HAL process (Silicon One Hardware Abstraction Layer), forcing the switch to reload. That means an attacker does not need to maintain persistence — they can achieve denial of service by simply triggering the crash. One exploit attempt, two outcomes: root compromise or network outage, both available on demand.
CVSS 9.8 puts this alongside Log4Shell and Spring4Shell. Unlike those application-layer flaws where filtering or sanitizing traffic could catch an attack above the vulnerability, this is the device doing the routing — there is no layer above it to save you.
This Is Your AI Data Center Fabric
This matters more than a typical switch CVE because of what Silicon One is. Cisco launched the Silicon One G300 in February 2026 specifically as the networking backbone for AI compute at scale: 102.4 Tbps switching capacity, designed for hyperscalers, neoclouds, sovereign clouds, and enterprise GPU clusters running training and inference workloads. The affected Nexus 9000 models are not legacy infrastructure — they are the switches physically routing GPU-to-GPU traffic in production AI data centers today.
Root RCE on one of these switches does not just mean an attacker owns a network device. It means they can observe, intercept, or disrupt every data flow between your GPU nodes. At the infrastructure layer, there is no application-level defense to fall back on.
Are You Affected?
Ten specific Nexus 9000 models with Silicon One ASICs are affected across 45 NX-OS releases from version 10.3(1) through 10.6(3s). Run this command on your switch to identify your hardware:
show moduleCheck the PID output against the affected model list in Cisco’s official security advisory. Confirmed affected models include the N9K-C9804, N9K-C9808, N9364E-SG2-Q, N9364E-SG2-O, N9336C-SE1, and five others in the N9324C and N9348Y families.
Three categories are explicitly not affected: Nexus 9000 switches running in ACI mode, the Nexus 3000 series, and the Nexus 7000 series. If you are running standalone NX-OS on a Silicon One model, you are in scope.
What to Do Now
The permanent fix is a NX-OS software upgrade. Use Cisco’s Software Checker to identify the earliest fixed release for your specific hardware and version. If you cannot patch immediately, two mitigations buy time:
Option 1 — iACL: Apply an infrastructure access control list to block TCP traffic to locally configured switch addresses on ports 43210 and 43211. Allow only required management and control-plane traffic. Test this in your environment before deploying — Cisco validated the approach in a lab, but every network design is different.
Option 2 — Live Protect Shield: Cisco released shield lp00031 for NX-OS 10.6(3), which automatically blocks exploitation attempts while you arrange the full upgrade. Check version compatibility first — this does not cover all 45 affected releases.
Both options are mitigations, not solutions. Patch anyway.
The Window Is Open — For Now
Cisco PSIRT confirmed on September 3 that no active exploitation or public proof-of-concept code is known. That status will not hold indefinitely. CVSS 9.8 zero-auth RCE flaws routinely attract researcher attention within days of public disclosure. The infrastructure at risk — AI training fabric at hyperscale — is high-value enough that motivated attackers will not wait for you to catch up. Patch now, not after the next incident report lands in your inbox. Full technical details are available at The Hacker News and Security Affairs.
