LightGBM versions up to 4.7.0 fail to validate child and split array values when parsing text models, allowing attackers to craft malicious model files that trigger out‑of‑bounds writes in the leaf_depth_ buffer during SHAP prediction. This out‑of‑bounds write (CWE‑787) corrupts memory and could lead to arbitrary code execution, a system crash, or data tampering if utilized by an attacker.

The LightGBM library from lightgbm‑org is affected. All releases through 4.7.0, including 4.6.0 and earlier, are vulnerable.

The vulnerability has a CVSS score of 8.5, indicating high severity, while the EPSS score of less than 1% suggests a low but non‑zero exploitation probability. It is not listed in CISA’s KEV catalog. The attack vector is likely local or remote if an attacker can supply a malicious model file to the SHAP prediction function; no privileged escalation is required, but the model must be executed in a process with sufficient memory to exploit the buffer overrun.