Keras has an untrusted deserialization vulnerability

高 7.8CVE-2026-1462 · GHSA-4f3f-g24h-fr8m · PYSEC-2026-2547 · 13/04/2026 公開 · 15/07/2026 更新

要点

CVE-2026-1462 は keras, keras (keras), openshift ai (redhat) に影響する深刻度「高」 (CVSS 7.8) の脆弱性です。
今後30日間に悪用される確率（EPSS）：0.40%（全脆弱性の 34.7% より高い）。
対処：3.13.2, 2.25.7 以降に更新してください。

パッケージ	影響を受けるバージョン	修正済みバージョン
keras PyPI	`< 3.13.2`	`3.13.2`
keras keras	`= 3.13.0`	修正版なし
openshift ai redhat	`≥ 2.25, < 2.25.7`	`2.25.7`

概要

A vulnerability in the TFSMLayer class of the keras package, version 3.13.0, allows attacker-controlled TensorFlow SavedModels to be loaded during deserialization of .keras models, even when safe_mode=True. This bypasses the security guarantees of safe_mode and enables arbitrary attacker-controlled code execution during model inference under the victim's privileges. The issue arises due to the unconditional loading of external SavedModels, serialization of attacker-controlled file paths, and the lack of validation in the from_config() method.

NVD の説明

A vulnerability in the `TFSMLayer` class of the `keras` package, version 3.13.0, allows attacker-controlled TensorFlow SavedModels to be loaded during deserialization of `.keras` models, even when `safe_mode=True`. This bypasses the security guarantees of `safe_mode` and enables arbitrary attacker-controlled code execution during model inference under the victim's privileges. The issue arises due to the unconditional loading of external SavedModels, serialization of attacker-controlled file paths, and the lack of validation in the `from_config()` method.