CVE-2026-12570 is a resource exhaustion vulnerability in keras-team/keras (<= 3.15.0) where the H5IOStore.__getitem__ method in keras/src/saving/saving_lib.py does not validate the shape or size of datasets when loading .keras model files via keras.models.load_model(). This allows an attacker to craft malicious model files that cause unbounded memory allocation, leading to out-of-memory conditions and process termination (exit code 137). The flaw bypasses the prior fix for CVE-2026-0897, which addressed a similar issue in a different component. The attack vector includes poisoned models from public repositories or malicious model registries, posing a risk to ML pipelines that load untrusted models.

Successful exploitation results in denial of service due to out-of-memory conditions causing the process to terminate unexpectedly. There is no impact on confidentiality or integrity reported. The vulnerability affects availability of the affected application when loading malicious model files.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, avoid loading untrusted or unverified .keras model files to mitigate risk.