CVE-2026-12570

Denial of Service via HDF5 Shape Bomb in keras.models.load_model() in keras-team/keras

Published Aug 10, 2026 CVSS 3.1 · 5.0 Medium EPSS < 1% KEV · Not listed
Description
A vulnerability in keras-team/keras versions <= 3.15.0 allows for a denial of service (DoS) attack when loading malicious .keras model files via the keras.models.load_model() function. The H5IOStore.__getitem__ method in keras/src/saving/saving_lib.py does not validate the shape or size of datasets, leading to unbounded memory allocation. A specially crafted .keras file can exploit this flaw to trigger an out-of-memory (OOM) condition, causing the process to be terminated (exit code 137). This issue bypasses the fix for CVE-2026-0897, which only addressed a similar vulnerability in KerasFileEditor. The attack vector includes poisoned models from public repositories or malicious model registries, posing a risk to machine learning pipelines that process untrusted models.

Analysis

AI-generated

Technical impact

Keras versions up to 3.15.0 contain a flaw in the file loading routine that fails to bound the size of H5 datasets. A specially crafted .keras file can trigger an unbounded memory allocation during the load, causing an out‑of‑memory condition and terminating the process. The weakness corresponds to CWE‑770, which describes uncontrolled resource consumption leading to denial of service.

Affected Systems

The vulnerable package is keras by the Keras Team. All releases with a version number less than or equal to 3.15.0 are affected, including the latest 3.15.0 release. No additional affected products are listed.

Risk and Exploitability

The CVSS score of 5.0 indicates a moderate severity for a DoS scenario. The EPSS score of 0.00128 indicates a very low probability of exploitation. The vulnerability is not listed in the CISA KEV catalog. The attack vector is likely to involve a malicious .keras file obtained from public repositories or a compromised model registry; the attacker must supply the file to an environment that executes keras.models.load_model(). If executed, the process will be killed with exit code 137, disrupting any machine‑learning pipelines that rely on that process.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade Keras to a version newer than 3.15.0, which contains the proper bounds check on H5 dataset sizes.
  • For environments that must load external models, restrict the trusted source list and manually vet model files before loading.
  • Run the model loading routine in a sandbox or container with memory limits to contain any OOM condition that the vulnerability could trigger.

Generated by OpenCVE AI on August 11, 2026 at 02:05 UTC.

Tracking

Sign in to view the affected projects.

Advisories

Source ID Title
Github GHSA Github GHSA GHSA-74m6-m3xx-3vmj Keras model loading is vulnerable to denial of service through HDF5 shape bombs
History
Type Values Removed Values Added
First Time appeared Keras-team
Keras-team keras
Vendors & Products Keras-team
Keras-team keras
Type Values Removed Values Added
References
Metrics threat_severity

None

cvssV3_1

{'score': 5.0, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:N/I:N/A:H'}

threat_severity

Moderate

Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'poc', 'Technical Impact': 'partial'}, 'version': '2.0.3'}

Type Values Removed Values Added
Description A vulnerability in keras-team/keras versions <= 3.15.0 allows for a denial of service (DoS) attack when loading malicious .keras model files via the keras.models.load_model() function. The H5IOStore.__getitem__ method in keras/src/saving/saving_lib.py does not validate the shape or size of datasets, leading to unbounded memory allocation. A specially crafted .keras file can exploit this flaw to trigger an out-of-memory (OOM) condition, causing the process to be terminated (exit code 137). This issue bypasses the fix for CVE-2026-0897, which only addressed a similar vulnerability in KerasFileEditor. The attack vector includes poisoned models from public repositories or malicious model registries, posing a risk to machine learning pipelines that process untrusted models.
Title Denial of Service via HDF5 Shape Bomb in keras.models.load_model() in keras-team/keras
Weaknesses CWE-770
References
Metrics cvssV3_0

{'score': 5.5, 'vector': 'CVSS:3.0/AV:L/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H'}

Weaknesses
  • CWE-770

    Allocation of Resources Without Limits or Throttling

Sources

Status
PUBLISHED
Assigner
@huntr_ai
Published
Updated
Aug 10, 2026, 5:37 PM
Reserved
Jun 18, 2026, 1:55 AM
Status
Awaiting Analysis
Published
Aug 10, 2026, 7:16 AM
Modified
Oct 23, 2026, 8:10 PM
Severity
Moderate
Public date
Aug 10, 2026, 6:29 AM