CVE-2025-9906

Source
https://nvd.nist.gov/vuln/detail/CVE-2025-9906
Import Source
https://storage.googleapis.com/cve-osv-conversion/osv-output/CVE-2025-9906.json
JSON Data
https://api.osv.dev/v1/vulns/CVE-2025-9906
Aliases
Downstream
Published
2025-09-19T09:15:36Z
Modified
2025-10-14T17:57:08.366139Z
Severity
  • 7.3 (High) CVSS_V3 - CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H CVSS Calculator
Summary
[none]
Details

The Keras Model.loadmodel method can be exploited to achieve arbitrary code execution, even with safemode=True.

One can create a specially crafted .keras model archive that, when loaded via Model.loadmodel, 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.enableunsafedeserialization() 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.enableunsafe_deserialization() needs to appear first in the archive and the Lambda with arbitrary code needs to be second.

References

Affected packages

Git / github.com/keras-team/keras

Affected ranges

Type
GIT
Repo
https://github.com/keras-team/keras
Events

Affected versions

v3.*

v3.0.0
v3.0.1
v3.0.2
v3.0.3
v3.0.4
v3.0.5
v3.1.0
v3.1.1
v3.10.0
v3.2.0
v3.2.1
v3.3.0
v3.3.1
v3.3.2
v3.3.3
v3.4.0
v3.4.1
v3.5.0
v3.6.0
v3.7.0
v3.8.0
v3.9.0