---
id: "CVE-2026-65119"
kind: "vulnerability"
aliases: ["CVE-2026-65119"]
title: "NVIDIA TensorRT-LLM: KvCacheAwareRouter tokenizer loading can run attacker code"
severity: "high"
cvss: 8.4
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-tooling", "llm-serving", "nvidia", "python", "tensorrt-llm"]
not_affected_if: ["router-not-used"]
published_at: "2026-10-27T13:16:59+00:00"
first_seen_at: "2026-10-27T13:31:51+00:00"
updated_at: "2026-10-27T13:32:21+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-65119", "first_seen_at": "2026-10-27T13:31:51+00:00"}]
---

# NVIDIA TensorRT-LLM: KvCacheAwareRouter tokenizer loading can run attacker code

NVIDIA TensorRT-LLM loads a tokenizer in its KvCacheAwareRouter from a source an attacker can influence, pulling in code from an untrusted control sphere. A successful exploit can lead to code execution, information disclosure, data tampering, or denial of service.

_Summary, check and fix written by a model from the sources; labelled as such._

## How to check

- Check the installed TensorRT-LLM version with `pip show tensorrt_llm` and compare it to the fixed version in the NVIDIA advisory.
- Check whether the KvCacheAwareRouter is enabled in your serving configuration; the sources do not say which settings enable it.

## Fix

- Upgrade TensorRT-LLM to the patched release named in the NVIDIA security bulletin (version unknown from the sources).
- Until patched, restrict which tokenizer paths or model repositories the router can load from; the sources do not give a specific workaround.

## Timeline

- 2026-10-27T13:31:51+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-65119

## From the sources

NVIDIA TensorRT-LLM for any platform contains a vulnerability in KvCacheAwareRouter tokenizer loading, where an attacker could cause inclusion of functionality from an untrusted control sphere. A successful exploit might lead to code execution, information disclosure, data tampering, and denial of service.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_