PatchSiren

PatchSiren cyber security CVE debrief

CVE-2026-12484 keras-team CVE debrief

A vulnerability in keras-team/keras version 3.15.0 allows unsafe deserialization of attacker-controlled PyTorch pickle data through the public `keras.layers.TorchModuleWrapper.from_config` method. This method invokes `torch.load(..., weights_only=False)` without requiring an explicit unsafe opt-in, such as a `safe_mode=False` parameter. When called outside a `SafeModeScope(True)` context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. The vulnerability arises because the method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode. Affected product deployments should be reviewed for potential exposure.

Vendor
keras-team
Product
keras-team/keras
CVSS
HIGH 7.8
CISA KEV
Not listed in stored evidence
Original CVE published
2026-07-19
Original CVE updated
2026-07-19
Advisory published
2026-07-19
Advisory updated
2026-07-19

Who should care

Users of keras-team/keras version 3.15.0 who process untrusted Keras layer configurations using the `keras.layers.TorchModuleWrapper.from_config` method should be aware of this vulnerability. Affected operator, platform, vulnerability-management, and security-team impact should be considered when assessing potential exposure.

Technical summary

The vulnerability arises because the `keras.layers.TorchModuleWrapper.from_config` method does not enforce safe deserialization practices unless explicitly guarded by Keras safe mode. This method invokes `torch.load(..., weights_only=False)` without requiring an explicit unsafe opt-in. When called outside a `SafeModeScope(True)` context, the absence of an ambient safe mode state permits unsafe deserialization by default. This issue can lead to arbitrary code execution if untrusted Keras layer configurations are processed using this method. Affected product context and defensive impact should be considered when assessing potential exposure.

Defensive priority

High priority should be given to updating keras-team/keras to a version that enforces safe deserialization practices or applying compensating controls to restrict access to the affected method.

Recommended defensive actions

  • Update keras-team/keras to a version that enforces safe deserialization practices
  • Apply compensating controls to restrict access to the affected method
  • Monitor for and respond to potential exploitation attempts
  • Review affected product deployments in managed environments and assign an owner for follow-up
  • Confirm whether exposure is present through normal change control where exposure is confirmed
  • Check relevant monitoring, detection, and logs for exposed assets that need extra review
  • Track exceptions, retest remediated assets, and close the item only after evidence is documented

Evidence notes

Evidence is limited to a single source reference from Huntr. Further verification is needed to confirm the vulnerability and its impact. Additional review of related configurations and potential exposures is recommended. Defenders should verify affected product deployments and assess potential operational impact.

Official resources

AI-assisted PatchSiren debrief based on the supplied source corpus. The CVE record was published on 2026-07-19T20:16:28.800Z and has not been modified since then.