NVIDIA Triton Inference Server: Crafted input causes an out-of-bounds write in the server

Impact

Crafted input causes an out-of-bounds write in the server. On a shared inference tier this is a noisy-neighbour weapon: one tenant's request kills the server process and takes every co-resident model with it, and the GPU sits idle until the pod restarts.

Who can reach it

Network. Anyone who can reach the Triton HTTP or gRPC endpoint. In most clusters that is anything on the pod network; where ingress is loosely scoped it is the internet. No authentication step exists in Triton itself to stop it.

What to do

Roll to the fixed Triton container image per bulletin 5691. Cost: an ordinary rolling deployment restart - no driver, firmware or node change. Worth pairing with an audit of Triton endpoint exposure, since almost every bug in this component is only interesting because the endpoint is reachable.

References

Related entries

• NVIDIA Triton Inference Server: An attacker who can locate and reach the Python backend's shared memory region corruptsCVE-2025-23329 · NVIDIA Triton Inference ServerHigh
• NVIDIA Triton Inference Server: An invalid request drives an excessive memory allocation and a segmentation faultCVE-2025-23331 · NVIDIA Triton Inference ServerHigh
• NVIDIA Triton Inference Server: Extra-large payloads trigger an improper-exceptional-condition check and crashCVE-2025-33201 · NVIDIA Triton Inference ServerHigh
• NVIDIA Triton Inference Server: Extra-large payloads cause a stack overflow and kill the serverCVE-2025-33202 · NVIDIA Triton Inference ServerHigh
• NVIDIA Triton Inference Server: Improper validation of a specified quantity in input crashes the serverCVE-2025-33211 · NVIDIA Triton Inference ServerHigh
• NVIDIA Triton Inference Server: A large compressed payload - a decompression bomb against the HTTP endpointCVE-2026-24158 · NVIDIA Triton Inference ServerHigh