HighVulnerability

CVE-2026-12482: keras-team/keras tar extraction allows symlinks outside extraction directory

Source: NVD(opens in a new tab)
Identifier: CVE-2026-12482
Published: July 14, 2026
Record updated: October 9, 2026

Affected: keras < 3.12.3, fixed in 3.12.3
keras >= 3.13.0, < 3.15.0, fixed in 3.15.0
Known exploitation: Not listed in the CISA Known Exploited Vulnerabilities catalog at the last check.
EPSS: 0.3%

Summary

CVE-2026-12482 affects keras-team/keras version 3.12.0. A crafted tar archive can bypass the `filter_safe_tarinfos` validation in `keras/src/utils/file_utils.py`, because symlink entries skip the `is_path_in_dir` check applied to regular files. Symlinks can then be created outside the extraction directory, enabling symlink-based file read, file overwrite, or directory escape. The source says the impact is greatest on Python 3.10 and 3.11, where this filter is the sole defense against tar path traversal.

Mitigation

The source does not state a fix yet. Check the original advisory for updates.

Details

Scores and exploitation

Patch available: Yes
EPSS: 0.3%Estimated probability that this CVE is exploited in the next 30 days, from the FIRST Exploit Prediction Scoring System. Read on October 14, 2026.
Disclosure date: July 14, 2026

Weaknesses and techniques

Weakness (CWE): CWE-22
Attack pattern (CAPEC): CAPEC-126

Affected

Packages: keras@< 3.12.3 (fixed: 3.12.3)
keras@>= 3.13.0, < 3.15.0 (fixed: 3.15.0)

Classification

Attack type: Other
Attack sophistication: Moderate
AI component targeted: Framework
Impact: Confidentiality,
Integrity
Labels: security

How this record was produced

Summaries, mitigations and labels on this page are generated automatically from the source. The linked source is authoritative. The source gives this advisory a description and no title, so the title above is a headline generated from the source text.

Original source: https://nvd.nist.gov/vuln/detail/CVE-2026-12482

First tracked: July 14, 2026 at 08:08 AM ET, from NVD/CVE Database

Classifier: claude-haiku-5-5, prompt v4

Classifier confidence: 90% (self-assessed)

Affected packages in the Exposure Registry

Matched by package name and ecosystem. Each entry shows whether the package delegates to a language model and how many tracked packages depend on it.

kerasPyPINo LLM dependency detected · 6 tracked dependents

Summary

Mitigation

Affected packages in the Exposure Registry

Related items