AIKIDO-2026-574328
keras is vulnerable to Unsafe Deserialization
78
High Risk
This Affects:
kerasTL;DR
TorchModuleWrapper.from_config restores a base64-encoded torch.nn.Module with torch.load(..., weights_only=False). Its safe-mode check only blocked when ambient in_safe_mode() was truthy, so a direct from_config call—or any path where the deserialization scope is unset—still unpickled attacker-controlled data and could execute arbitrary code. The fix treats an unset safe mode as safe and refuses the load unless safe_mode is explicitly disabled.
Who does this affect?
You are affected if you are using a version that falls within the vulnerable range and you deserialize TorchModuleWrapper configs or models that contain them from untrusted sources.
Background info
keras is vulnerable to Unsafe Deserialization in versions 3.11.3 - 3.14.1.
How to fix this
Upgrade the keras library to the patch version.
Are You Affected?
Connect your repositories to instantly see whether vulnerable or malicious packages exist in your codebase.
Free. No credit card required.
I consent to receiving marketing communications based on Aikido’s Privacy Policy.
SOC 2Compliant
ISO 27001Compliant