GPU VulnDB

Database/Firmware, BMC & network fabric

Arista EOS BGP agent: crafted packet crashes BGP, possibly code execution on the switch

CVSS 9.3CVE-2026-121289Firmware, BMC & network fabriccurated

Impact

A crafted packet makes the BGP agent on an Arista EOS switch terminate, which drops every BGP session on that device and can disrupt routing until the agent recovers. Arista adds that a sophisticated attacker might also get arbitrary code execution on the switch. Many GPU clusters run their Ethernet AI and RoCE fabrics as BGP-routed leaf-spine, so if a leaf or spine loses all its sessions at once, every GPU node behind it can drop out of a training job. Code execution on a fabric switch would also put traffic from every tenant crossing it at risk.

Who can reach it

Over the network against a switch running BGP with an established peering session. The advisory says the vulnerable code runs whenever BGP is enabled and a session is up. It does not say whether the attacker must be a BGP peer or only needs to reach the device. Only switches on the multi-agent routing model are exposed; devices on ribd or not running BGP are not.

What to do

Upgrade EOS to a fixed release: 4.37.0F, 4.36.3M, 4.35.7M, 4.34.9M or 4.33.11M or later in each train. End-of-support trains from 4.25.1F onward are affected, and Arista lists no hotfix. Plan it as a switch image upgrade per the EOS upgrade procedure, and drain traffic off each leaf or spine in turn. Until you can upgrade: on sessions that do not use the Tunnel Encapsulation Attribute, apply neighbor default received attribute discard 23, and restrict peering with TCP-AO authentication. To check exposure, run show ip bgp summary and show running-config all | inc service routing protocols.

References

Related entries

All Firmware, BMC & network fabric entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.