This vulnerability in keras-team/keras 3.14.0 arises from the DiskIOStore.make method's failure to properly sanitize layer names used in directory path construction. Although forward slashes are restricted, directory traversal components like '..' are not filtered, permitting attackers to craft malicious model files that cause the software to write or create files outside the designated temporary working directory. This improper limitation of pathname to a restricted directory (CWE-22) can lead to unauthorized file system access during model save or load operations.

An attacker who can supply or influence layer names in a Keras model can exploit this vulnerability to perform unauthorized file system operations, including creating directories or writing files outside the intended temporary directory. This may lead to partial compromise of confidentiality, integrity, and availability of the system where the model is processed. The CVSS 3.0 score of 6.1 reflects a medium severity with local attack vector, low attack complexity, no privileges required, user interaction required, scope change, and low impact on confidentiality, integrity, and availability.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is released, avoid loading or saving untrusted Keras models that may contain malicious layer names. Monitor vendor communications for updates regarding patches or official mitigations.