CVE-2026-16497: 
Triton Inference Server vulnerability analysis and mitigation

CVE-2026-16497 is an Excessive Iteration vulnerability in NVIDIA Triton Inference Server for Linux that allows a remote, unauthenticated attacker to cause a denial of service condition. The vulnerability affects Triton Inference Server versions up to and including 26.06. It was published on September 8, 2026, and is part of NVIDIA Security Bulletin 5875 (August 2026). The CVSS v3.1 base score is 7.5 (High) (Feedly, NVIDIA Product Security).

The root cause is classified as CWE-834 (Excessive Iteration), meaning the server can be induced into performing an abnormally large number of loop iterations when processing attacker-controlled input. The attack vector is network-based, requires no authentication, no user interaction, and low attack complexity, making it trivially automatable. No public technical write-ups or proof-of-concept code have been identified at this time (Feedly, NVIDIA Product Security).

Successful exploitation leads exclusively to a denial of service — there is no confidentiality or integrity impact. An attacker can exhaust server resources by triggering excessive iteration, potentially rendering the Triton Inference Server unavailable and disrupting AI/ML inference workloads that depend on it. Given that Triton is commonly deployed in production ML serving pipelines, availability disruption could have downstream effects on dependent applications and services (Feedly).

No proof-of-concept code or active in-the-wild exploitation has been reported as of the time of this report. The vulnerability is not listed in the CISA Known Exploited Vulnerabilities (KEV) catalog. The EPSS score is approximately 0.43%, indicating a low probability of exploitation in the near term. However, NVD's SSVC assessment notes the vulnerability is automatable, meaning exploitation could be scripted without manual interaction (Feedly).

NVIDIA has addressed this vulnerability in Triton Inference Server versions after 26.06 as part of Security Bulletin 5875. Users should upgrade to the latest available version of NVIDIA Triton Inference Server. As a network-based, unauthenticated vulnerability, network-level controls such as restricting access to the Triton inference endpoint to trusted clients only can reduce exposure until patching is complete (NVIDIA Product Security).

Coverage has been limited to automated vulnerability tracking platforms such as VulDB and SecurityOnline, with no notable researcher commentary or significant community discussion identified at this time (SecurityOnline, VulDB).