CVE-2025-9906
- CVSS Vector: CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H
- Published: 2025-09-19T09:15:36.353
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.
Related CVE by CWE
No related CWE found.
Top CVE for Vendor
No vendor taxonomy on this entry.
Recently Exploited Similar Vulnerabilities
No recent KEV-listed items for this vendor/product.
How to fix CVE-2025-9906
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 […]
Exploit Difficulty: HARD
⏱️ Time to exploit: > 4 hours
🛠️ Required skills: Advanced security expertise
💰 Public exploits: Rare or not public
How to Fix:
- Check if you're running the affected product
- Update to the latest patched version
- If patching is not immediately possible: restrict network exposure, apply least-privilege access
- Test the fix in a staging environment first
- Review logs for signs of exploitation
- Monitor for IOCs (Indicators of Compromise)
- Enable automatic security updates
- Set up vulnerability monitoring
- Review and harden security configurations
Exploit Difficulty Assessment
Vulnerability Timeline
CVE details first published to NVD database
Added to this CVE tracking system
Detection Rules & IOCs
No specific detection rules generated for this vulnerability type.
No vendor/product data available.