Triton Inference Server CVE-2026-47477

NVIDIA Triton Inference Server for Linux contains a remotely exploitable denial-of-service vulnerability (CVE-2026-47477) that could disrupt AI/ML inference operations for organizations relying on Triton for production workloads. Unauthenticated attackers can trigger the flaw without network privileges, and no vendor patch is currently available, requiring immediate interim controls to mitigate availability risk.

Within 24 hours, create an inventory of all systems running NVIDIA Triton Inference Server and assess their network exposure; isolate or air-gap production instances where feasible. Within 7 days, implement network-based compensating controls including strict firewall rules limiting Triton access to trusted internal sources only, enable rate limiting at the network perimeter, and deploy a web application firewall or load balancer with denial-of-service protections configured to detect and block malformed requests. Within 30 days, monitor NVIDIA security advisories and the NVD for patch availability; schedule immediate testing and deployment of the vendor patch upon release, prioritizing production Triton instances.

Network segmentation to restrict inbound access to Triton services; implement rate limiting and input validation at the network perimeter; deploy denial-of-service detection and mitigation (load balancer with DoS protections, WAF rules, or network-based IPS); restrict Triton API access to trusted internal sources only; monitor system resource consumption for signs of exploitation attempts.