Arbitrary Code execution in Keras Safe Mode
Vulnerability 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.
Impact Analysis
Refer to official advisory for detailed impact metrics.
Remediation
Ensure systems are updated to the latest vendor-supplied patch levels.
Am I Vulnerable?
Launch our assessment wizard to check if your infrastructure is exposed to • CVE-2025-9906
Credits & Attribution
The following person or organization is credited with identifying this vulnerability, as recorded in the NVD database:
- Gabriele Digregorio
More from Keras-team
View All →Affected Vendor
Keras-team
View all reports →