CVE-2026-86288
Received Received - Intake

Out-of-Bounds Read in GPTQModel Triton Dequantization Kernel

Publication date: 2026-09-07

Last updated on: 2026-09-07

Assigner: VulDB

Description
A vulnerability has been found in ModelCloud GPTQModel up to 7.2.0. This vulnerability affects unknown code of the file gptqmodel/nn_modules/qlinear/tritonv2.py of the component Triton dequantization kernel. Such manipulation of the argument g_idx leads to out-of-bounds read. The attack can be executed remotely. The exploit has been disclosed to the public and may be used. Upgrading to version 7.3.0 is able to resolve this issue. The name of the patch is 877c732f7d7dccd56a729844c6a5bd20f3aa8bb1. Upgrading the affected component is recommended.

CVSS Scores

EPSS Scores

Probability:
Percentile:

Meta Information

Published
2026-09-07
Last Modified
2026-09-07
Generated
2026-09-07
AI Q&A
2026-09-07
EPSS Evaluated
N/A
NVD
CVE-2026-86288
EUVD
EUVD-2026-72303

Affected Vendors & Products

Showing 3 associated CPEs
Vendor Product Version / Range
modelcloud gptqmodel to 7.2.0 (inc)
modelcloud gptqmodel 7.3.0
modelcloud gptqmodel 7.2.0

Helpful Resources

Exploitability

CWE
CWE Icon
KEV
KEV Icon
CWE ID Description
CWE-125 The product reads data past the end, or before the beginning, of the intended buffer.
CWE-119 The product performs operations on a memory buffer, but it reads from or writes to a memory location outside the buffer's intended boundary. This may result in read or write operations on unexpected memory locations that could be linked to other variables, data structures, or internal program data.

Attack-Flow Graph

AI Quick Actions

Instant insights powered by AI
Executive Summary

CVE-2026-86288 is an out-of-bounds read vulnerability in ModelCloud GPTQModel's Triton dequantization kernel. It occurs when processing a malicious GPTQ checkpoint with activation reordering. The issue arises because the kernel uses a crafted g_idx value to index scales and qzeros buffers without validating if it falls within the valid range [0, num_groups). This can lead to reading out-of-bounds memory, causing CUDA context crashes or exposing adjacent GPU memory.

Detection Guidance

Detecting this vulnerability requires checking if your GPTQModel version is vulnerable (<=7.2.0) and inspecting model checkpoints for malicious g_idx values. Use pip show GPTQModel to check the installed version. For model files, inspect the g_idx tensor in the checkpoint for values outside the valid range [0, num_groups). No direct network detection commands are available as this is a local model loading issue.

Impact Analysis

This vulnerability allows remote attackers to crash the CUDA context (denial of service) by tricking victims into loading a malicious model file. It may also expose adjacent GPU memory, potentially leaking sensitive data. The impact is limited to systems using Triton dequantization with untrusted model files.

Compliance Impact

This vulnerability does not directly affect compliance with GDPR or HIPAA as it involves an out-of-bounds read in a machine learning model quantization library. However, if exploited, it could lead to denial-of-service or memory exposure, which may indirectly impact systems handling sensitive data. No specific compliance impact is documented in the provided context.

Mitigation Strategies

Upgrade GPTQModel to version 7.3.0 or later immediately. If upgrading is not possible, avoid loading untrusted GPTQ checkpoints with desc_act=True enabled. Validate g_idx values in checkpoints before loading by checking if all values are within [0, num_groups). Disable Triton acceleration if not strictly necessary.

Chat Assistant

Ask questions about this CVE
Hi! I’m here to help you understand CVE-2026-86288. Ask me anything about the vulnerability, its impact, or mitigation strategies.
0/70

EPSS Chart