high CVSS 7.3 EPSS 0.21%

CVE-2025-9905·Python vulnerability

The Keras Model.load_model method can be exploited to achieve arbitrary code execution, even with safe_mode=True. One can create a specially crafted .h5/.hdf5 model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed. This is achieved by crafting a special .h5 archive file that uses the Lambda layer feature of keras which allows arbitrary Python code in the form of pickled code. The vulnerability comes from the fact that the safe_mode=True option is not honored when reading .h5 archives. Note that the .h5/.hdf5 format is a legacy format supported by Keras 3 for backwards compatibility.

Severity
high
Software
Python
Fixed in
3.11.3
Published

How likely is exploitation

FIRST's Exploit Prediction Scoring System puts the probability of this CVE being exploited in the next 30 days at 0.21% — higher than 11% of all published CVEs. EPSS is an independent forecast of exploitation, not a severity rating: it says nothing about how much damage a successful attack does, which is what a CVSS score measures.

Read the two together. CVE-2025-9905 scores 7.3 on CVSS, so a successful attack would be serious — but at 0.21%, almost nobody is currently attempting it. That is a different situation from a flaw under mass exploitation, and it is usually the difference between patching on your next maintenance window and patching tonight.

Affected versions

From: 3.0.0

Until: 3.11.3

Fixed in: 3.11.3

How to fix this CVE

Update Keras to version 3.11.3 or later to resolve unsafe deserialization of legacy .h5 model files. Users should avoid loading untrusted .h5/.hdf5 model archives, as the safe_mode parameter does not prevent arbitrary code execution through pickled Lambda layer payloads. Verify your Keras installation and apply the patch immediately if you load model files from external sources.

sudo dnf check-update python3-keras && sudo dnf update python3-keras

Defensia detects this vulnerability

How to check if you are affected

  1. Run 'pip show keras' or 'python3 -c "import keras; print(keras.__version__)"' to check the installed Keras version
  2. Review your application code for calls to keras.models.load_model() with .h5 or .hdf5 file paths
  3. Audit file upload endpoints or model serving pipelines that accept .h5 files from external sources
  4. After patching, re-run 'python3 -c "import keras; print(keras.__version__)"' to confirm version 3.11.3 or later is installed

FAQ

What is CVE-2025-9905?

A deserialization vulnerability in Keras 3.0.0–3.11.2 where the Model.load_model() function fails to enforce safe_mode restrictions when loading legacy .h5 model archives, allowing attackers to execute arbitrary Python code via crafted pickled Lambda layers.

Is CVE-2025-9905 being actively exploited?

No, CVE-2025-9905 is not listed on the CISA Known Exploited Vulnerabilities catalog and no public exploits have been disclosed.

What versions of Python are affected by CVE-2025-9905?

Keras versions 3.0.0 through 3.11.2 are vulnerable; Python itself is not directly affected. Any system running Keras within this version range that loads .h5 model files is at risk.

How do I check if my server is vulnerable to CVE-2025-9905?

Run 'python3 -c "import keras; print(keras.__version__)"' and compare the output to version 3.11.3; if your version is 3.11.2 or earlier, you are vulnerable.

Does Defensia detect CVE-2025-9905?

Yes — Defensia's CVE advisory scanner compares installed package versions against the NVD database. If Keras is installed on a monitored server, CVE-2025-9905 will appear in your dashboard with remediation steps.

Related Python CVEs

CVE-2026-28505CVSS 10Tautulli is a Python based monitoring and tracking tool for Plex Media Server. Prior to version 2.17.0, the str_eval() function in notification_handler.py implements a sandboxed eval() for notification text templates. The sandbox attempts to restrict callable names by inspecting code.co_names of the compiled code object. However, co_names only contains names from the outer code object. When a lambda expression is used, it creates a nested code object whose attribute accesses are stored in code.co_consts, NOT in code.co_names. The sandbox never inspects nested code objects. This issue has been patched in version 2.17.0.
CVE-2026-34938CVSS 10PraisonAI is a multi-agent teams system. Prior to version 1.5.90, execute_code() in praisonai-agents runs attacker-controlled Python inside a three-layer sandbox that can be fully bypassed by passing a str subclass with an overridden startswith() method to the _safe_getattr wrapper, achieving arbitrary OS command execution on the host. This issue has been patched in version 1.5.90.
CVE-2026-33054CVSS 10Mesop is a Python-based UI framework that allows users to build web applications. Versions 1.2.2 and below contain a Path Traversal vulnerability that allows any user supplying an untrusted state_token through the UI stream payload to arbitrarily target files on the disk under the standard file-based runtime backend. This can result in application denial of service (via crash loops when reading non-msgpack target files as configurations), or arbitrary file manipulation. This vulnerability heavily exposes systems hosted utilizing FileStateSessionBackend. Unauthorized malicious actors could interact with arbitrary payloads overwriting or explicitly removing underlying service resources natively outside the application bounds. This issue has been fixed in version 1.2.3.
CVE-2026-25632CVSS 10EPyT-Flow is a Python package designed for the easy generation of hydraulic and water quality scenario data of water distribution networks. Prior to 0.16.1, EPyT-Flow’s REST API parses attacker-controlled JSON request bodies using a custom deserializer (my_load_from_json) that supports a type field. When type is present, the deserializer dynamically imports an attacker-specified module/class and instantiates it with attacker-supplied arguments. This allows invoking dangerous classes such as subprocess.Popen, which can lead to OS command execution during JSON parsing. This also affects the loading of JSON files. This vulnerability is fixed in 0.16.1.
CVE-2026-9135CVSS 9.9IBM Langflow OSS 1.0.0 through 1.10.0 Langflow versions up to 1.9.2 (commit 94981c443d4918517b9e8163d70fc598dc33a32d) contain a code injection vulnerability in the Policies component's ToolGuard integration that bypasses the allow_custom_components=false security control. The vulnerability exists because the validation mechanism only checks the main component source code in node_template["code"]["value"] but fails to validate dynamic CodeInput fields that store generated ToolGuard Python files. Attackers can embed malicious Python code in these unvalidated dynamic fields, which are persisted in Flow.data and later executed server-side when a guarded tool is invoked through the ToolGuard runtime. This allows authenticated users with flow creation privileges to achieve arbitrary Python code execution on the backend despite custom component restrictions. The vulnerability can be escalated through cross-tenant flow manipulation via the agentic MCP update_flow_component_field tool, which accepts attacker-controlled user_id parameters, enabling attackers to inject malicious code into victim users' flows. When combined with publicly accessible flows and specific misconfigurations (AUTO_LOGIN=true, NEW_USER_IS_ACTIVE=true), the attack can be conducted with reduced authentication requirements.

References

Track CVEs across your fleet automatically

Defensia scans your Linux servers and tells you exactly which ones are running vulnerable versions — including CVE-2025-9905. Free for 1 server.

Get started free