The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.

Note: This report has already been discussed with the Google OSS VRP team, who recommended that I reach out directly to the Keras team. I’ve chosen to do so privately rather than opening a public issue, due to the potential security implications. I also attempted to use the email address listed i...

Keras is vulnerable to Deserialization of Untrusted Data

Arbitrary Code Execution in Keras Keras versions prior to 3.11.0 allow for arbitrary code execution when loading a crafted .keras model archive, even when safemode=True. The issue arises because the archive’s config.json is parsed before layer deserialization. This can invoke...

GHSA-36FQ-JGMW-4R9C Keras is vulnerable to Deserialization of Untrusted Data

Arbitrary Code Execution in Keras Keras versions prior to 3.11.0 allow for arbitrary code execution when loading a crafted .keras model archive, even when safemode=True. The issue arises because the archive’s config.json is parsed before layer deserialization. This can invoke...

CVE-2025-9906

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...

PYSEC-2025-76

The Keras Model.loadmodelmethod can be exploited to achieve arbitrary code execution, even with safemode=True.One can create a specially crafted .kerasmodel archive that, when loaded via Model.loadmodel, will trigger arbitrary code to be executed. This is achieved by crafting a special config.jso...

PYSEC-2025-123

The Keras Model.loadmodelmethod can be exploited to achieve arbitrary code execution, even with safemode=True.One can create a specially crafted .h5/.hdf5model archive that, when loaded via Model.loadmodel, will trigger arbitrary code to be executed.This is achieved by crafting a special .h5archi...

UBUNTU-CVE-2025-9905

The Keras Model.loadmodel method can be exploited to achieve arbitrary code execution, even with safemode=True. One can create a specially crafted .h5/.hdf5 model archive that, when loaded via Model.loadmodel, will trigger arbitrary code to be executed. This is achieved by crafting a special...

Improper Control of Dynamically-Managed Code Resources

Overview keras is a Keras is a high-level neural networks API for Python.. Affected versions of this package are vulnerable to Improper Control of Dynamically-Managed Code Resources via the Model.loadmodel method. An attacker can execute arbitrary code by providing a specially crafted .h5 or .hdf...

CVE-2025-9905

CVE-2025-9905 affects Keras Model.load_model for .h5/.hdf5 archives, where safe_mode=True is ignored and Lambda layer can trigger arbitrary code execution. Vulnerability details: loading a crafted .h5 can execute code via pickled Lambda; legacy .h5 format is involved. Remediation: upgrade Keras t...

CVE-2025-9905 Arbitary Code execution in Keras load_model()

The Keras Model.loadmodel method can be exploited to achieve arbitrary code execution, even with safemode=True. One can create a specially crafted .h5/.hdf5 model archive that, when loaded via Model.loadmodel, will trigger arbitrary code to be executed. This is achieved by crafting a special...

CVE-2025-9905

The Keras Model.loadmodel method can be exploited to achieve arbitrary code execution, even with safemode=True. One can create a specially crafted .h5/.hdf5 model archive that, when loaded via Model.loadmodel, will trigger arbitrary code to be executed. This is achieved by crafting a special...

Linux Distros Unpatched Vulnerability : CVE-2025-1550

The Linux/Unix host has one or more packages installed that are impacted by a vulnerability without a vendor supplied patch available. - The Keras Model.loadmodel function permits arbitrary code execution, even with safemode=True, through a manually constructed, malicious .keras archive. By...

Linux Distros Unpatched Vulnerability : CVE-2025-8747

The Linux/Unix host has one or more packages installed that are impacted by a vulnerability without a vendor supplied patch available. - A safe mode bypass vulnerability in the Model.loadmodel method in Keras versions 3.0.0 through 3.10.0 allows an attacker to achieve arbitrary code execution by...

Linux Distros Unpatched Vulnerability : CVE-2020-14212

The Linux/Unix host has one or more packages installed that are impacted by a vulnerability without a vendor supplied patch available. - FFmpeg through 4.3 has a heap-based buffer overflow in aviogetstr in libavformat/aviobuf.c because dnnbackendnative.c calls ffdnnloadmodelnative and a certain...

CVE-2025-8747

A safe mode bypass vulnerability in the Model.loadmodel method in Keras versions 3.0.0 through 3.10.0 allows an attacker to achieve arbitrary code execution by convincing a user to load a specially crafted .keras model archive...

Duplicate Advisory: Keras safe mode bypass vulnerability

Duplicate Advisory This advisory has been withdrawn because it is a duplicate of GHSA-c9rc-mg46-23w3. This link is maintained to preserve external references. Original Description A safe mode bypass vulnerability in the Model.loadmodel method in Keras versions 3.0.0 through 3.10.0 allows an...

PYSEC-2025-75

A safe mode bypass vulnerability in the Model.loadmodel method in Keras versions 3.0.0 through 3.10.0 allows an attacker to achieve arbitrary code execution by convincing a user to load a specially crafted .keras model archive...

PYSEC-2025-75

A safe mode bypass vulnerability in the Model.loadmodel method in Keras versions 3.0.0 through 3.10.0 allows an attacker to achieve arbitrary code execution by convincing a user to load a specially crafted .keras model archive...

Deserialization of Untrusted Data

Overview keras is a Keras is a high-level neural networks API for Python.. Affected versions of this package are vulnerable to Deserialization of Untrusted Data via the function savinglib.loadmodel, which identifies the model source file, directory, or Hugging Face repository and then calls the...

CVE-2025-8747

CVE-2025-8747 corresponds to a safe-mode bypass in Keras Model.load_model, allowing arbitrary code execution by loading a crafted .keras archive. Connected IBM bulletins confirm the vulnerability affects Keras 3.0.0–3.10.0 and describe a bypass via manipulated config.json or inner Lambda mechanis...