The vulnerability in NVIDIA Triton Inference Server arises from insufficient input validation coupled with the ability to request a large number of outputs. This weakness can cause the server process to crash, preventing it from responding to legitimate inference requests. The core issue aligns with CWE‑789, leading to a denial of service without impacting data confidentiality or integrity.

All versions of NVIDIA Triton Inference Server may be affected. The CNA list only includes the vendor/product; no specific version information is provided. Users should verify their deployment against NVIDIA documentation or support channels.

The CVSS score of 7.5 indicates a high risk, while the EPSS score is unavailable, making it hard to gauge exploitation frequency. The vulnerability is not in the CISA KEV catalog. Based on the description, the likely attack vector involves sending a crafted inference request with a large payload or many outputs to exhaust resources and trigger a crash. Precise prerequisites or attack vector details are inferred from the description.