GHSA-77wq-646f-jrm2: Duplicate Advisory: The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.

1. GitHub Advisory Database
2. GitHub Reviewed
3. GHSA-77wq-646f-jrm2

Duplicate Advisory: The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.

High severity GitHub Reviewed Published Sep 19, 2025 to the GitHub Advisory Database • Updated Sep 19, 2025

Withdrawn This advisory was withdrawn on Sep 19, 2025

Affected versions

>= 3.0.0, < 3.11.3

Duplicate Advisory

This advisory has been withdrawn because it is a duplicate of GHSA-36rr-ww3j-vrjv. This link is maintained to preserve external references.

Original 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 .h5/.hdf5 model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed.

This is achieved by crafting a special .h5 archive file that uses the Lambda layer feature of keras which allows arbitrary Python code in the form of pickled code. The vulnerability comes from the fact that the safe_mode=True option is not honored when reading .h5 archives.

Note that the .h5/.hdf5 format is a legacy format supported by Keras 3 for backwards compatibility.

References

• GHSA-36rr-ww3j-vrjv
• https://nvd.nist.gov/vuln/detail/CVE-2025-9905
• keras-team/keras#21602

Published by the National Vulnerability Database

Sep 19, 2025

Published to the GitHub Advisory Database

Sep 19, 2025

Last updated

Sep 19, 2025