---
id: CVE-2026-84286
title: >-
  ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a
  CUDA kernel without first verifying that the index is within bounds
summary: >-
  ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a
  CUDA kernel without first verifying that the index is within bounds. A
  malicious model can force the code to read and invoke an out‑of‑range function
  pointer, re…
severity: none
published: '2026-10-12'
updated: '2026-10-12'
sourceUpdated: '2026-10-12T18:17:24.160'
source: NVD
sourceUrl: 'https://nvd.nist.gov/vuln/detail/CVE-2026-84286'
references:
  - url: 'https://github.com/turboderp-org/exllamav3/pull/310'
    label: cret@cert.org
  - url: 'https://kb.cert.org/vuls/id/369611'
    label: cret@cert.org
  - url: 'https://www.kb.cert.org/vuls/id/369611'
    label: af854a3a-2127-422b-91ae-364da2661108
tags:
  - nvd
ingestedAt: '2026-10-12T18:00:08.218Z'
---

## Overview

ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.

## Remediation

Refer to the linked advisories for vendor-supplied fixes and affected version ranges.