Security Bulletin
Summary
Keras is used by many machine learning frameworks and applications as part of their deep learning infrastructure. Remote attackers can execute arbitrary code, leading to full system compromise, data breaches, and potential lateral movement where the identified vulnerability is present. Critical vulnerability in Keras model loading may allow remote code execution.
Vulnerability Details
CVEID: CVE-2025-1550
DESCRIPTION: The Keras Model.load_model function permits arbitrary code execution, even with safe_mode=True, through a manually constructed, malicious .keras archive. By altering the config.json file within the archive, an attacker can specify arbitrary Python modules and functions, along with their arguments, to be loaded and executed during model loading.
CWE: CWE-94: Improper Control of Generation of Code ('Code Injection')
CVSS Source: NVD
CVSS Base score: 9.8
CVSS Vector: (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H)
CVEID: CVE-2025-8747
DESCRIPTION: A safe mode bypass vulnerability in the `Model.load_model` method in Keras versions 3.0.0 through 3.10.0 allows an attacker to achieve arbitrary code execution by convincing a user to load a specially crafted `.keras` model archive.
CWE: CWE-502: Deserialization of Untrusted Data
CVSS Source: NVD
CVSS Base score: 7.8
CVSS Vector: (CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H)
CVEID: CVE-2025-9906
DESCRIPTION: 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 .keras model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed. This is achieved by crafting a special config.json (a file within the .keras archive) that will invoke keras.config.enable_unsafe_deserialization() to disable safe mode. Once safe mode is disable, one can use the Lambda layer feature of keras, which allows arbitrary Python code in the form of pickled code. Both can appear in the same archive. Simply the keras.config.enable_unsafe_deserialization() needs to appear first in the archive and the Lambda with arbitrary code needs to be second.
CWE: CWE-502: Deserialization of Untrusted Data
CVSS Source: NVD
CVSS Base score: 7.3
CVSS Vector: (CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H)
Affected Products and Versions
| Affected Product(s) | Version(s) |
| WML-CPD | 4.0 to 5.2.2 releases |
Remediation/Fixes
Vulnerability is fixed in RT25.1. Customers are not to use RT 24.1, IBM strongly recommends addressing the vulnerability immediately by upgrading their runtime to 25.1 to mitigate risk.
RT25.1 is currently available in version 5.2.2. Guide to set up Keras on RT25.1 can be found here
Workarounds and Mitigations
None
Get Notified about Future Security Bulletins
References
Acknowledgement
Change History
19 Dec 2025: Initial Publication
*The CVSS Environment Score is customer environment specific and will ultimately impact the Overall CVSS Score. Customers can evaluate the impact of this vulnerability in their environments by accessing the links in the Reference section of this Security Bulletin.
Disclaimer
According to the Forum of Incident Response and Security Teams (FIRST), the Common Vulnerability Scoring System (CVSS) is an "industry open standard designed to convey vulnerability severity and help to determine urgency and priority of response." IBM PROVIDES THE CVSS SCORES ""AS IS"" WITHOUT WARRANTY OF ANY KIND, INCLUDING THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. CUSTOMERS ARE RESPONSIBLE FOR ASSESSING THE IMPACT OF ANY ACTUAL OR POTENTIAL SECURITY VULNERABILITY. In addition to other efforts to address potential vulnerabilities, IBM periodically updates the record of components contained in our product offerings. As part of that effort, if IBM identifies previously unidentified packages in a product/service inventory, we address relevant vulnerabilities regardless of CVE date. Inclusion of an older CVEID does not demonstrate that the referenced product has been used by IBM since that date, nor that IBM was aware of a vulnerability as of that date. We are making clients aware of relevant vulnerabilities as we become aware of them. "Affected Products and Versions" referenced in IBM Security Bulletins are intended to be only products and versions that are supported by IBM and have not passed their end-of-support or warranty date. Thus, failure to reference unsupported or extended-support products and versions in this Security Bulletin does not constitute a determination by IBM that they are unaffected by the vulnerability. Reference to one or more unsupported versions in this Security Bulletin shall not create an obligation for IBM to provide fixes for any unsupported or extended-support products or versions.
Document Location
Worldwide
Was this topic helpful?
Document Information
Modified date:
19 December 2025
Initial Publish date:
19 December 2025
UID
ibm17255407