AdvisoriesToolsStatsChecklistSubmitGet the email
← All advisories
CVE-2026-65119highCode execution2026-10-27

CVE-2026-65119, CVE-2026-65120 and CVE-2026-24269: NVIDIA TensorRT-LLM tokenizer loading and SSRF bugs, update to 1.3.0rc25

NVIDIA's October 2026 TensorRT-LLM bulletin fixes three bugs in 1.3.0rc24. KvCacheAwareRouter's tokenizer loading can pull in functionality from an untrusted source, which can lead to code execution, and two server-side request forgery flaws can disclose or tamper with data. NVIDIA lists 1.3.0rc25 as the updated version.

01.

What it is

CVE-2026-65119 (CVSS 8.4, CWE-829): a flaw in KvCacheAwareRouter tokenizer loading lets an attacker cause inclusion of functionality from an untrusted control sphere, which might lead to code execution, information disclosure, data tampering and denial of service.

CVE-2026-65120 (CVSS 7.1, CWE-918): a server-side request forgery that might lead to information disclosure.

CVE-2026-24269 (CVSS 7.1, CWE-918): a server-side request forgery that might lead to information disclosure and data tampering.

All three carry a local attack vector (AV:L) in NVIDIA's scoring. NVIDIA doesn't publish more technical detail.

02.

How to check

  1. Check your TensorRT-LLM version. NVIDIA lists 1.3.0rc24 as affected.
  2. Check whether you use the KvCacheAwareRouter and load tokenizers from paths or repos you don't control.
03.

How to fix

  1. Update TensorRT-LLM to 1.3.0rc25 or later from the NVIDIA/TensorRT-LLM repo or package.
  2. Only load tokenizers and models from sources you trust.
pip show tensorrt-llm | grep -i version
04.

Sources

The weekly email

One email a week. Only what changed.

Every new security hole in AI coding tools, MCP servers, app builders and the stacks they generate (Next.js, Supabase, Node, PHP), each with how to check your app and the fix. No roundups of things you already read.

Free, one email a week. Read the latest issue first if you like. One-click unsubscribe in every email. We never sell or share addresses.