CVE-2026-92786: LightGBM through 4.7.0 Out-of-Bounds Write via Crafted Model
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 leafdepth buffer during feature contribution computation.
Affected Software
Event History
Frequently Asked Questions
What conditions are required to trigger the out-of-bounds write?
An attacker must cause LightGBM to parse a crafted text model containing invalid child or split array values, then have SHAP feature-contribution prediction performed using that model. The attack vector is local and requires user interaction, but no privileges are required.
Which deployments are most exposed?
Deployments that load text-format models from untrusted or insufficiently validated sources and perform SHAP prediction are exposed. The provided information does not establish whether default configurations load untrusted models or invoke feature-contribution computation.
How can I determine whether my environment is affected?
Check whether Microsoft LightGBM is version 4.7.0 or earlier and whether your workflow loads externally supplied text models. Also identify uses of SHAP or feature-contribution prediction, as the reported write occurs during that computation.