This vulnerability (CVE-2026-0897) affects Google Keras versions 3.0.0 through 3.13.0 on all platforms. The issue lies in the HDF5 weight loading component where resource allocation is not limited or throttled. An attacker can supply a specially crafted .keras archive containing a model.weights.h5 file with a dataset declaring an extremely large shape. This causes excessive memory allocation leading to exhaustion and a crash of the Python interpreter, resulting in a denial of service condition. The CVSS 4.0 base score is 7.1, indicating high severity. There are no known exploits in the wild. Red Hat has published advisories referencing this CVE, but no explicit patch or fix details are provided in the advisory content.

Successful exploitation allows a remote attacker to cause a denial of service by exhausting system memory and crashing the Python interpreter running Keras. This disrupts availability of applications using the affected Keras versions. There is no indication of privilege escalation, data disclosure, or code execution from the provided information.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Red Hat has published advisories referencing this vulnerability but does not explicitly state that a fix is available. Users should monitor official Google Keras and Red Hat advisories for patches or updates addressing this issue. Until a patch is available, avoid loading untrusted .keras archives containing model.weights.h5 files with potentially large datasets.