CVE-2026-92786

Overview

LightGBM through version 4.7.0 fails to properly validate child and split array values when parsing text model files. Processing a maliciously crafted model file during SHAP value prediction triggers an out-of-bounds memory write in the leaf_depth_ buffer. This vulnerability can lead to application instability, memory corruption, or potential code execution in the application context.

Description

LightGBM through 4.7.0 fails to validate child and split array values when parsing text models, allowing attackers to write out-of-bounds memory during SHAP prediction. Attackers can craft malicious model files with invalid node references that trigger out-of-bounds writes at attacker-chosen offsets in the leaf_depth_ buffer during feature contribution computation.

Impact

Impacts data science workflows, services, and machine learning pipelines that ingest untrusted LightGBM text models or calculate SHAP feature contributions. Exploitation undermines application integrity and availability, with potential confidentiality risks stemming from process memory manipulation.

Remediation

Upgrade LightGBM to a patched release newer than version 4.7.0. As a temporary control, ensure that model files ingested by applications originate exclusively from trusted, authenticated sources and avoid parsing untrusted model artifacts.

Risk context

Carries a high severity rating with a CVSS v4 score of 8.5 and CVSS v3 score of 7.8. Although EPSS data is not available, organizations ingesting third-party or user-supplied machine learning models should consider this high urgency.

Affected products

• Microsoft LightGBM <= 4.7.0