Arbitrary Code Execution in Keras via Lambda Layer Deserialization

Executive Summary

This vulnerability exists in keras-team/keras version 3.14.0 and allows for arbitrary code execution. It occurs because the function responsible for handling deserialization in the Lambda layer, specifically _raise_for_lambda_deserialization(), does not properly enforce a safe-mode guard when the safe_mode parameter is set to None. Since None is the default value when from_config() is called outside a SafeModeScope context, this logic error treats None (unset/default-deny) the same as False (explicitly disabled), effectively bypassing the safety check.

As a result, attacker-controlled marshal bytecode can be deserialized, which can lead to arbitrary operating system-level code execution within the server or user process. The vulnerability affects calls to keras.layers.deserialize(config), keras.models.clone_model(model), and any direct invocation of Lambda.from_config(config) without an enclosing SafeModeScope(True).

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can have severe impacts because it allows an attacker to execute arbitrary code at the operating system level within the context of the server or user process running the vulnerable Keras version. This means an attacker could potentially take control of the affected system, manipulate data, disrupt services, or use the compromised system as a foothold for further attacks.

Useful 0 Not Useful 0

Mitigation Strategies

To mitigate this vulnerability, ensure that any deserialization of the Lambda layer in keras is done within a SafeModeScope context with safe_mode explicitly set to True. Avoid calling Lambda.from_config(config) or keras.layers.deserialize(config) without this safe mode guard.

Additionally, consider updating or patching keras to a version where this vulnerability is fixed, or apply any available security patches from the maintainers.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability allows for arbitrary OS-level code execution due to improper deserialization in the Keras Lambda layer. Such a security flaw can lead to unauthorized access, data breaches, or manipulation of sensitive information.

Because of the potential for unauthorized access and data compromise, this vulnerability could negatively impact compliance with common standards and regulations such as GDPR and HIPAA, which require protection of personal and sensitive data against unauthorized access and breaches.

However, specific impacts on compliance depend on the context in which the vulnerable software is used and whether the vulnerability is exploited.

Useful 0 Not Useful 0