Sign in
Deutsch English Español Polski
Medium

Keras model loading is vulnerable to denial of service through HDF5 shape bombs

  • Disclosed: 10.08.2026
  • Updated: 11.10.2026 15:18

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.

Affected software

Click a technology to see all vulnerabilities affecting it.

PlatformPackageVulnerable versionsFixed in
Python (PyPI) keras < 3.15.0 3.15.0

References

CVSS score 5.5/10
CVSS vector
CVSS:3.0/AV:L/AC:L/PR:N/UI:R/S:U/C:N/I:N/A:H
Weakness (CWE)
CWE-770
Platform
Python (PyPI)
Package
keras
Disclosed
10.08.2026
Updated
11.10.2026 15:18

Sources and licenses

OSV / GitHub Advisory Database - PYSEC-2026-3856 - PyPA Advisory Database; CC BY 4.0

OSV / GitHub Advisory Database - GHSA-74m6-m3xx-3vmj - CC BY 4.0

Vulnerability history: keras

All published vulnerabilities of this package (Python (PyPI)), newest first. Shows whether this is a one-off slip or a package that needs watching.

All (20) in the archive
Total entries
20
First → latest
04.2024 08.2026
Without a fix
1
Exploited in attacks
0

Severity breakdown

  • Critical 2
  • High 11
  • Medium 6
  • Low 1

Disclosures per year

Severity:
  1. Keras model loading is vulnerable to denial of service through HDF5 shape bombs this entry
    Medium · 5.5 CVE-2026-12570 < 3.15.0 3.15.0
  2. Keras: TorchModuleWrapper can deserialize unsafe PyTorch pickle data
    High · 7.8 CVE-2026-12484 < 3.12.3, >= 3.13.0, < 3.15.0 3.12.3, 3.15.0
  3. Keras: Lambda deserialization can bypass safe mode and execute code
    High · 8.8 CVE-2026-12481 < 3.12.3, >= 3.13.0, < 3.15.0 3.12.3, 3.15.0
  4. Keras versions prior to 3.14.0 are vulnerable to a path traversal issue in the archive extraction utilities located in `keras/src/utils/file_utils.py`. The functions `filter_safe_tarinfos()` and `filt
    High · 8.1 CVE-2026-11816 < 3.14.0 3.14.0
  5. Keras has an untrusted deserialization vulnerability
    High · 8.8 CVE-2026-1462 < 3.13.2 3.13.2
  6. Keras has a Local File Disclosure via HDF5 External Storage During Keras Weight Loading
    High · 7.1 CVE-2026-1669 >= 3.13.0, < 3.13.2, >= 3.0.0, < 3.12.1 3.13.2, 3.12.1
  7. Allocation of Resources Without Limits or Throttling in the HDF5 weight loading component in Google Keras 3.0.0 through 3.13.0 on all platforms allows a remote attacker to cause a Denial of Service (D
    High · 7.5 CVE-2026-0897 >= 3.0.0, < 3.13.1 3.13.1
  8. Keras Directory Traversal Vulnerability
    High · 9.8 CVE-2025-12060 < 3.12.0 3.12.0
  9. Keras framework vulnerable to deserialization of untrusted data
    Critical · 9.8 CVE-2025-49655 >= 3.11.0, < 3.11.3 3.11.3
  10. Keras is vulnerable to Deserialization of Untrusted Data
    High · 7.3 CVE-2025-9906 < 3.11.0 3.11.0
  11. The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.
    High CVE-2025-9905 >= 3.0.0, < 3.11.3 3.11.3
  12. Keras vulnerable to CVE-2025-1550 bypass via reuse of internal functionality
    High · 8.8 CVE-2025-8747 >= 3.0.0, < 3.11.0 3.11.0
  13. Arbitrary Code Execution via Crafted Keras Config for Model Loading
    High CVE-2025-1550 >= 3.0.0, < 3.9.0 3.9.0
  14. Keras code injection vulnerability
    Critical · 9.8 CVE-2024-3660 < 2.13.1rc0 2.13.1rc0

Know which vulnerabilities affect your application

With the DockRay integration and software inventory enabled, DockRay checks reported component versions against known vulnerabilities and notifies you about confirmed matches. Coverage depends on the supported technology and available advisory data.

Connect your application
DockRay newsletter

Application security in your inbox

Important vulnerabilities, actionable guidance and practical advice for your applications. Choose your technologies and follow what matters to your project.

  • A daily digest of new vulnerabilities, critical ones included.
  • Notifications about important updates.
  • Practical advice on application security and monitoring.

Leave empty to receive updates for every technology.

Privacy policy

Free to subscribe. Unsubscribe with one click in any email.