CVE-2025-9906
The Keras Model.load_model method can be exploited to achieve arbitrary code execution, even with safe_mode=True.
Does this matter?
High impact if exploited, but EPSS currently rates exploitation as unlikely (0.20%). Schedule it in the normal patch cycle and watch for a rise in EPSS or a public exploit.
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 4.0
- 8.6 HIGHCVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/M
- EPSS
- 0.20% probability · 10th percentile
- CISA KEV
- Not listed
- Weakness
- CWE-502
- Affected
- keras/keras
- Source
- cve-coordination@google.com
References
- https://github.com/keras-team/keras/pull/21429Issue Tracking
Source: NVD record, EPSS from FIRST.org, KEV from CISA. Refreshed daily.