UTC

keras

20 published advisories; version 3.15.0 or later fixes all but 1 of them.

Registry
PyPI
Advisories
20
Latest fix
3.15.0
Exploited in the wild
None listed
Advisories with no fix yet

One advisory has no fixed release in any version yet. What can be done meanwhile:

General guidance from ShieldPi, not a vendor fix

  • Path traversal. Limit the folders it can reach at the operating-system level, such as a container or a user with access to only those folders, and keep credentials and keys outside them.GHSA-cjgq-5qmw-rcj6

Check your projects

shieldpi agent-config checks agent and MCP packages from npm and PyPI, not keras. Any scanner that reads the OSV database finds it in your lockfiles, for example Google's open-source OSV-Scanner.

osv-scanner scan source -r .

Or check a lockfile against ShieldMap in your browser.

Malicious lookalikes

Packages reported malicious by OpenSSF whose names imitate keras. Check that you install keras itself.

All AI lookalikes →
Provenance

Source: github.com/keras-team/keras, named in the package metadata (not verified)

0 of the last 10 releases of keras were published with build provenance: a signed record of the repository and commit each was built from, which the registry checks before it accepts the release.

  • 3.12.4
  • 3.15.1
  • 3.12.3
  • 3.15.0
  • 3.12.2
  • 3.14.1
  • 3.14.0
  • 3.12.1
  • 3.13.2
  • 3.13.1

Green: published with provenance. Grey: without.

OpenSSF Scorecard: 7.4 of 10 on 2026-11-18.

No dangerous CI workflows
10/10
CI tokens read-only by default
0/10
Build dependencies pinned
5/10
Branch protection
0/10
Changes reviewed before merge
10/10
Signed releases
Not checked
Actively maintained
10/10
Security policy
10/10
Full Scorecard →

Provenance from deps.dev (CC BY 4.0); Scorecard from OpenSSF (CDLA-Permissive-2.0). As of 2026-11-18.

Advisories
AdvisorySeverityAffectedFixed inPublished
Keras model loading is vulnerable to denial of service through HDF5 shape bombsCVE-2026-12570Medium< 3.15.03.15.02026-08-10
Keras: HDF5 links can disclose local file contentsCVE-2026-9335Medium< 3.12.33.12.32026-08-02
Keras: HDF5 links can disclose local file contentsCVE-2026-9335Medium>= 3.13.0, < 3.15.03.15.02026-08-02
Keras: TorchModuleWrapper can deserialize unsafe PyTorch pickle dataCVE-2026-12484High< 3.12.33.12.32026-07-19
Keras: TorchModuleWrapper can deserialize unsafe PyTorch pickle dataCVE-2026-12484High>= 3.13.0, < 3.15.03.15.02026-07-19
Keras: tar extraction permits symlink-based path traversalCVE-2026-12482Low< 3.12.33.12.32026-07-14
Keras: tar extraction permits symlink-based path traversalCVE-2026-12482Low>= 3.13.0, < 3.15.03.15.02026-07-14
Keras: Lambda deserialization can bypass safe mode and execute codeCVE-2026-12481High< 3.12.33.12.32026-07-03
Keras: Lambda deserialization can bypass safe mode and execute codeCVE-2026-12481High>= 3.13.0, < 3.15.03.15.02026-07-03
Keras: HDF5 virtual datasets can disclose local filesCVE-2026-12480Medium< 3.12.33.12.32026-07-01
Keras: HDF5 virtual datasets can disclose local filesCVE-2026-12480Medium>= 3.13.0, < 3.15.03.15.02026-07-01
Keras: DiskIOStore permits path traversal through crafted layer namesCVE-2026-12479Medium< 3.12.33.12.32026-06-22
Keras: DiskIOStore permits path traversal through crafted layer namesCVE-2026-12479Medium>= 3.13.0, < 3.15.03.15.02026-06-22
Keras archive extraction utilities allow path traversal and arbitrary file writesCVE-2026-11816High< 3.14.03.14.02026-06-11
Keras vulnerable to DoS via Malicious .keras Model (HDF5 Shape Bomb Causes Petabyte Allocation in KerasFileEditor)CVE-2026-0897High>= 3.0.0, <= 3.12.03.12.12026-05-06
Keras vulnerable to DoS via Malicious .keras Model (HDF5 Shape Bomb Causes Petabyte Allocation in KerasFileEditor)CVE-2026-0897High>= 3.13.0, < 3.13.23.13.22026-05-06
Keras has an untrusted deserialization vulnerabilityCVE-2026-1462High< 3.13.23.13.22026-04-13
Keras has a Local File Disclosure via HDF5 External Storage During Keras Weight LoadingCVE-2026-1669High>= 3.0.0, < 3.12.13.12.12026-02-18
Keras has a Local File Disclosure via HDF5 External Storage During Keras Weight LoadingCVE-2026-1669High>= 3.13.0, < 3.13.23.13.22026-02-18
Keras Directory Traversal VulnerabilityCVE-2025-12060High<= 3.11.33.12.02025-12-02
Keras is vulnerable to arbitrary local file loading and Server-Side Request ForgeryCVE-2025-12058Medium< 3.12.03.12.02025-10-29
Keras framework vulnerable to deserialization of untrusted dataCVE-2025-49655Critical>= 3.11.0, < 3.11.33.11.32025-10-17
Keras is vulnerable to Deserialization of Untrusted DataCVE-2025-9906High< 3.11.03.11.02025-09-19
The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is lCVE-2025-9905High>= 3.0.0, < 3.11.33.11.32025-09-19
Keras vulnerable to CVE-2025-1550 bypass via reuse of internal functionalityCVE-2025-8747High>= 3.0.0, < 3.11.03.11.02025-08-12
Arbitrary Code Execution via Crafted Keras Config for Model LoadingCVE-2025-1550High>= 3.0.0, < 3.9.03.9.02025-03-11
keras Path Traversal vulnerabilityCVE-2024-55459Medium<= 3.7.0No fix yet2025-01-08
Keras code injection vulnerabilityCVE-2024-3660Critical< 2.13.1rc02.13.1rc02024-04-16

Sources: GitHub Advisory Database, reviewed and malware advisories (CC BY 4.0, https://github.com/advisories); advisories published by Cursor (https://github.com/cursor/cursor/security/advisories); advisories published by OpenClaw (https://github.com/openclaw/openclaw/security/advisories); advisories published by vLLM (https://github.com/vllm-project/vllm/security/advisories); advisories published by Open WebUI (https://github.com/open-webui/open-webui/security/advisories); advisories published by n8n (https://github.com/n8n-io/n8n/security/advisories); advisories published by Flowise (https://github.com/FlowiseAI/Flowise/security/advisories); advisories published by Langflow (https://github.com/langflow-ai/langflow/security/advisories); advisories published by the MCP Python SDK project (https://github.com/modelcontextprotocol/python-sdk/security/advisories); advisories published by MLflow (https://github.com/mlflow/mlflow/security/advisories); CISA Known Exploited Vulnerabilities catalog (CC0). Data as of 2026-11-22. ShieldMap describes how each problem works and how to fix it; it does not publish exploit code. Report a correction to [email protected].