CVE-2025-9906

GHSA-36fq-jgmw-4r9c HIGH
Published September 19, 2025

The Keras Model.load_model method can be exploited to achieve arbitrary code execution, even with safe_mode=True. One can create a specially crafted .keras model archive that, when loaded via...

Full analysis pending. Showing NVD description excerpt.

Affected Systems

Package Ecosystem Vulnerable Range Patched
keras pip < 3.11.0 3.11.0
keras pip No patch

Severity & Risk

CVSS 3.1
7.3 / 10
EPSS
0.1%
chance of exploitation in 30 days
KEV Status
Not in KEV
Sophistication
N/A

Recommended Action

Patch available

Update keras to version 3.11.0

Compliance Impact

Compliance analysis pending. Sign in for full compliance mapping when available.

Technical Details

NVD Description

The Keras Model.load_model method can be exploited to achieve arbitrary code execution, even with safe_mode=True. One can create a specially crafted .keras model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed. This is achieved by crafting a special config.json (a file within the .keras archive) that will invoke keras.config.enable_unsafe_deserialization() to disable safe mode. Once safe mode is disable, one can use the Lambda layer feature of keras, which allows arbitrary Python code in the form of pickled code. Both can appear in the same archive. Simply the keras.config.enable_unsafe_deserialization() needs to appear first in the archive and the Lambda with arbitrary code needs to be second.

CVSS Vector

CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H

Timeline

Published
September 19, 2025
Last Modified
September 23, 2025
First Seen
September 19, 2025