Cyber Resilience

CVE-2026-1669

Path Traversal in Keras 3.0.0 – 3.13.1

Published
11 February 2026
Modified
15 July 2026
CVSS Score v4 7.1
Click a component to see what it means
Raw vectorCVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:H/VI:L/VA:N/SC:N/SI:N/SA:N/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:X
EPSS Score 0.0030 22th percentile
Risk Priority 55 floored blend · peak EPSS

Summary

CVE-2026-1669 is a high-severity External Control of File Name or Path (CWE-73) vulnerability in Keras Keras. Its CVSS base score is 7.1 (High).

Operationally, exploitation aligns with the MITRE ATT&CK technique Data from Local System (T1005); ranked at the 22th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog.

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-4 (Information Flow Enforcement) — see the control section below for these in your framework.

Deeper analysis AI-assisted summary

Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.

CVE-2026-1669 is an arbitrary file read vulnerability in the model loading mechanism, specifically the HDF5 integration, affecting Keras versions 3.0.0 through 3.13.1 on all supported platforms. Published on 2026-02-11, the flaw enables a remote attacker to read local files and disclose sensitive information via a crafted .keras model file that utilizes HDF5 external dataset references. It is associated with CWEs-73 (External Control of File Name or Path) and CWE-200 (Exposure of Sensitive Information to an Unauthorized Actor), and carries a CVSS v3.1 base score of 7.5 (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N).

A remote attacker can exploit this vulnerability without privileges, over the network, with low attack complexity and no user interaction required. By tricking a victim into loading a malicious .keras model file—such as through shared repositories, downloads, or collaborative ML workflows—the attacker achieves arbitrary local file reads, potentially exposing sensitive data like configuration files, credentials, or proprietary datasets.

Mitigation details and further guidance are available in the advisory from Google Security Research at https://github.com/google/security-research/security/advisories.

OWASP Top 10 for Web (2025)

EU & UK References

Vulnerability Data

Arbitrary file read in the model loading mechanism (HDF5 integration) in Keras versions 3.0.0 through 3.13.1 on all supported platforms allows a remote attacker to read local files and disclose sensitive information via a crafted .keras model file utilizing HDF5…

more

external dataset references.

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

T1005 Data from Local System Collection
Adversaries may search local system sources, such as file systems, configuration files, local databases, virtual machine files, or process memory, to find files of interest and sensitive data prior to Exfiltration.
T1083 File and Directory Discovery Discovery
Adversaries may enumerate files and directories or may search in specific locations of a host or network share for certain information within a file system.
T1134.001 Token Impersonation/Theft Stealth
Adversaries may duplicate then impersonate another user's existing token to escalate privileges and bypass access controls.
T1552 Unsecured Credentials Credential Access
Adversaries may search compromised systems to find and obtain insecurely stored credentials.
T1552.001 Credentials In Files Credential Access
Adversaries may search local file systems and remote file shares for files containing insecurely stored credentials.
T1552.002 Credentials in Registry Credential Access
Adversaries may search the Registry on compromised systems for insecurely stored credentials.
Derived from this CVE’s CWE(s) via the direct CWE→ATT&CK cross-walk.

CVEs Like This One

CVE-2026-11816Same product: Keras Keras
CVE-2026-12482Same product: Keras Keras
CVE-2025-8747Same product: Keras Keras
CVE-2025-9906Same product: Keras Keras
CVE-2026-12481Same product: Keras Keras
CVE-2025-9905Same product: Keras Keras
CVE-2024-3660Same product: Keras Keras
CVE-2025-1550Same product: Keras Keras
CVE-2026-0897Same product: Keras Keras
CVE-2024-55459Same product: Keras Keras

Affected Assets

keras
keras
3.0.0 — 3.13.1

Mitigating Controls

Control response

Prevent
Stop it (NIST 800-53)

Detect
Catch it (NIST detect / respond)

Harden
Shrink the surface (DISA STIG)
  • 7 hardening rules · 6 OS baselines
Validate
Prove the fix (OWASP ASVS)
  • V10.4.9
  • V11.7.1
  • V14.1.2
  • V14.2.4

Mitigating Controls (NIST 800-53 r5) AI

Access enforcement directly stops unauthorized actors from obtaining sensitive information.

Information flow enforcement structurally prevents sensitive data from reaching unauthorized recipients.

Protection of information at rest prevents unauthorized exposure of stored sensitive data.

Transmission confidentiality mechanisms stop exposure of sensitive data on the wire.

Least privilege reduces the set of actors who can reach sensitive information.

Input validation directly rejects or sanitizes untrusted path strings before they reach filesystem operations.

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.

PR.PS-06 full match
prevents

Secure SDLC practices include input validation and path sanitization that eliminate this weakness.

PR.AA-05 mostly match
prevents

Least-privilege file authorization directly limits damage from externally controlled paths.

PR.DS-10 mostly match
prevents

PR.DS-10 mostly prevents CWE-200 by directly eliminating unauthorized access to sensitive data-in-use, yet only partially addresses the weakness because CWE-200 spans many other exposure vectors outside runtime protection.

PR.IR-01 mostly match
prevents

PR.IR-01's segmentation/zero-trust controls largely eliminate network-level unauthorized access paths that enable exposure, yet CWE-200 spans many additional vectors (API responses, logs, app logic) that network controls alone cannot close.

PR.AA-01 partial match
prevents

PR.AA-01 supplies proper credential lifecycle controls that reduce unauthorized access paths, yet leaves many other exposure vectors (error messages, logging, side channels, etc.) unaddressed.

PR.AA-03 partial match
prevents

Authentication verifies actor identity and is a prerequisite for access decisions, yet addresses only one facet of the broad set of exposure vectors in CWE-200.

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.

prevents

Restricting anonymous or unknown access and encrypting high-value information limits the exposure of sensitive data that would otherwise be obtainable by unauthorized actors.

prevents

Suppressing system details, error specifics, and previous log-on information until successful authentication reduces the information an unauthenticated attacker can gather.

prevents

By requiring owners to assign sensitivity labels and corresponding handling rules, the control ensures that information is not left unmarked and therefore reduces the chance that sensitive data will be exposed to unauthorized actors.

prevents

Requiring encryption, access controls, and recipient authentication for transfers directly reduces the chance that sensitive data reaches an unauthorized observer.

prevents

Secure delivery, protected storage, and confidentiality of allocation records limit exposure of authentication material to unauthorized observers.

prevents

Requiring defined procedures, assigned roles, and technical/organizational measures for handling PII reduces the chance that sensitive personal data will be exposed to unauthorized actors through inadequate handling or missing safeguards.

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).

Ubuntu 22.04 (1 rule)
  • V-260470 Ubuntu 22.04 LTS, when booted, must require authentication upon booting into single-user and maintenance modes. prevents CWE-200
Ubuntu 24.04 (2 rules)
  • V-270647 Ubuntu 24.04 LTS must not have the telnet package installed. prevents CWE-200
  • V-270675 Ubuntu 24.04 LTS when booted must require authentication upon booting into single-user and maintenance modes. prevents CWE-200
Windows 10 (1 rule)
  • V-220737 Administrative accounts must not be used with applications that access the Internet, such as web browsers, or with potential Internet sources, such as email. prevents CWE-200
Windows Server 2016 (1 rule)
  • V-224974 Domain-created Active Directory Organizational Unit (OU) objects must have proper access control permissions. prevents CWE-200
Windows Server 2019 (1 rule)
  • V-205743 Windows Server 2019 organization created Active Directory Organizational Unit (OU) objects must have proper access control permissions. prevents CWE-200
Windows Server 2022 (1 rule)
  • V-254395 Windows Server 2022 organization created Active Directory Organizational Unit (OU) objects must have proper access control permissions. prevents CWE-200

References