CVE-2025-9905
Keras 3.0.0 – 3.11.3
Raw vector
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/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:XSummary
CVE-2025-9905 is a high-severity Improper Control of Dynamically-Managed Code Resources (CWE-913) vulnerability in Keras Keras. Its CVSS base score is 7.3 (High).
Operationally, exploitation aligns with the MITRE ATT&CK technique Reflective Code Loading (T1620); ranked at the 10th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog; a public proof-of-concept is referenced.
This vulnerability is AI-related — categorised as Deep Learning Frameworks; in the Supply Chain and Deployment risk domain.
The strongest mitigations our analysis identified map to AC-3 (Access Enforcement) and AC-6 (Least Privilege) — see the control section below for these in your framework.
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2025-30277
Vulnerability Data
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…
more
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.
- CWE(s)
AI Security AnalysisAI
- AI Category
- Deep Learning Frameworks
- Risk Domain
- Supply Chain and Deployment
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: keras
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
- 2 hardening rules · 2 OS baselines
—
Mitigating Controls (NIST 800-53 r5) AI
Access enforcement directly stops unauthorized reads/writes to dynamic code resources by applying authorization checks at access time.
Least privilege reduces the set of subjects that can reach or modify dynamic code resources, limiting the weakness's reach.
Isolating security functions from non-security code prevents unintended manipulation of dynamically managed executable resources.
Process isolation keeps each process's dynamic code resources in separate domains, blocking cross-process tampering.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
Secure SDLC practices explicitly include controls that prevent improper handling of dynamic code resources.
Blocking unauthorized code execution directly limits the ability to abuse dynamically-managed resources.
Runtime-environment monitoring can detect exploitation of the weakness but does not prevent it.
Vulnerability identification can surface instances of CWE-913 but does not mitigate the root weakness.
Hardened configuration baselines can restrict dynamic code execution and variable access at runtime.
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
Security testing can detect dynamic code weaknesses but does not prevent them at design or coding time.
Secure development lifecycle mandates controls on dynamic code generation and resource management.
Application security requirements explicitly address restrictions on dynamic code execution and resource access.
Secure architecture principles require design controls that prevent improper dynamic code resource manipulation.
Secure coding standards directly prohibit unsafe dynamic code resource handling and injection patterns.
Environment separation reduces exposure of dynamic code resources but does not address the underlying weakness.
Hardening callouts derived
Configuration rules from DISA STIG baselines that bear on weaknesses of the type cited by this CVE. Each rule is shown with the relationship its mapping actually records, against the CWE it was authored against. Derived via CVE→CWE over `controls_xwalks` (authoritative rows only; rows rated `none` are excluded).
Windows 10 (1 rule)
- V-220726 Data Execution Prevention (DEP) must be configured to at least OptOut. prevents CWE-913
Windows 11 (1 rule)
- V-253283 Data Execution Prevention (DEP) must be configured to at least OptOut. prevents CWE-913