Connect your agent
AdvisoriesCriticalCheck your lockfilePluginHow it worksPricing Connect your agent

Advisory record · GHSA-36rr-ww3j-vrjv

The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.

CVSS 7.3 HighEPSS 0.0022 · 12th percentile

Published by the GitHub Advisory Database at 19 Sept 2025, 20:12 UTC. agentprime has not analysed this yet, so whether it affects any app is not known.

PackageEcosystemAffectedFixed in
kerasPyPI>=3.0.0, <3.11.33.11.3
Identifiers
GHSA-36rr-ww3j-vrjv · CVE-2025-9905 · PYSEC-2025-123
Weakness (CWE)
CWE-913
Tags
Published
19 Sept 2025
Updated
10 Sept 2026

Machine-written summary

Summary written by Claude Haiku from the source advisory. Not reviewed by a person.

The Keras package (PyPI keras, versions >=3.0.0 and <3.11.3) silently ignores safe_mode=True when Model.load_model is used on .h5 or .hdf5 files, so no safety check runs. An attacker who supplies a malicious model file that a victim loads, for example one distributed on a model hub, can run arbitrary code with the privileges of the Keras process, and no further interaction is needed. Upgrade keras to version 3.11.3 or later. Avoid loading .h5 or .hdf5 files from untrusted sources.

Written 5 Nov 2026.

Not analysed yet

We haven’t published our analysis of this Advisory yet. Once a person has reviewed it, this page will say what’s vulnerable, what has to be true for your app to be affected, and how to check.

Severity and exploitation

ScoreRatingCVSSVectorSource
7.3 High 4.0 CVSS:4.0/AV:L/AC:H/AT:P/PR:L/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H CNA (Google)
  • CISA KEV Not listed.
  • EPSS 0.0022, the 12th percentile, on 5 Nov 2026: FIRST’s estimate of the chance of exploitation in the next 30 days.

Timeline

  1. 31 Oct 2026, 14:34 UTC First seen by agentprime
  2. 31 Oct 2026, 14:34 UTC Confirmed by OSV

Sources

Facts on this page come from the GitHub Advisory Database (CC-BY 4.0), OSV and the other sources listed, as published. We don’t reproduce full advisory descriptions, which can contain exploit code.