Mend.io Vulnerability Database
The largest open source vulnerability database
What is a Vulnerability ID?
New vulnerability? Tell us about it!
CVE-2025-9906
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 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.
Affected Packages
keras (PYTHON):
Affected version(s) >=0.1.0 <3.11.0
Fix Suggestion:
Update to version 3.11.0
Do you need more information?
Contact Us
CVSS v4
Base Score:
8.6
Attack Vector
LOCAL
Attack Complexity
LOW
Attack Requirements
NONE
Privileges Required
LOW
User Interaction
PASSIVE
Vulnerable System Confidentiality
HIGH
Vulnerable System Integrity
HIGH
Vulnerable System Availability
HIGH
Subsequent System Confidentiality
HIGH
Subsequent System Integrity
HIGH
Subsequent System Availability
HIGH
CVSS v3
Base Score:
8.2
Attack Vector
LOCAL
Attack Complexity
LOW
Privileges Required
LOW
User Interaction
REQUIRED
Scope
CHANGED
Confidentiality
HIGH
Integrity
HIGH
Availability
HIGH
Weakness Type (CWE)
Deserialization of Untrusted Data
EPSS
Base Score:
0.08