A large compressed payload sent to the HTTP endpoint of NVIDIA Triton Inference Server can cause excessive memory or resource consumption, leading to the service becoming unresponsive and denying legitimate traffic. This resource exhaustion issue is classified under CWE-789 (Uncontrolled Resource Consumption). The result is a denial of service that can affect any application relying on the inference service.

The vulnerability affects NVIDIA Triton Inference Server. The affected products are those that expose the HTTP endpoint, and while specific vulnerable releases are not listed, it is inferred that any version of Triton that supports the HTTP API may be susceptible.

The issue carries a CVSS score of 7.5, indicating high severity, but the EPSS rate is below 1% and the vulnerability is not listed in the CISA KEV catalog, suggesting a low likelihood of exploitation at present. The attack vector is remote, via the publicly accessible HTTP interface; an attacker can send an oversized compressed request to trigger resource exhaustion. If exploited, the server may crash or become unresponsive, leading to service disruption until it is manually restarted or patched.