CVE-2025-9906 Arbitrary Code execution in Keras Safe Mode
Attackers can execute arbitrary code by loading a malicious Keras model archive that disables safe mode and exploits Lambda layers.
- Exploited in the wild
- Not known
- Login needed
- Any signed-in user
- Attackable
- Local or same network only
- Victim action needed
- A user has to click or open something
- Fix
- Available
What is CVE-2025-9906?
CVE-2025-9906 is a high-severity unsafe deserialization vulnerability in Keras-team Keras, published on 19 Sep 2025. It can only be exploited from the same machine or local network, and only if a user opens a file or follows a link.
Official 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.
Weakness: CWE-502
Which versions are affected by CVE-2025-9906?
3 open source packages ship an affected version, including keras. It affects 1 product.
| Package | Affected | Fixed in |
|---|---|---|
keras PyPI | All versions | None |
keras PyPI | All versions | 3.11.0 |
keras PyPI | All versions | 3.11.0 |
Products include Keras Keras.
How do I fix CVE-2025-9906?
Upgrade keras to 3.11.0 or later.
Upgrade, and only deserialise data from trusted sources or with an allow-list of types. More about Unsafe deserialization.
Is CVE-2025-9906 being exploited?
There is no known exploitation of CVE-2025-9906 in the wild and no public exploit code.
Other vulnerabilities in Keras-team Keras
- CVE-2024-3660 Arbitrary code injection vulnerability in Keras framework < 2.13
- CVE-2026-12481 Deserialization of Untrusted Data in keras-team/keras
- CVE-2026-1462 Safe Mode Bypass in keras-team/keras
- CVE-2025-8747 Keras safe_mode bypass allows arbitrary code execution when loading a malicious model.
- CVE-2026-11816 Path Traversal in keras-team/keras
Sources
Also known as AZL-67508, DEBIAN-CVE-2025-9906, GHSA-36fq-jgmw-4r9c, PYSEC-2025-76, UBUNTU-CVE-2025-9906.