Product:
Keras
Summary
Arbitrary file read in the model loading mechanism (HDF5 integration) in Keras allows a remote attacker to read local files and disclose sensitive information via a crafted .keras model file utilizing HDF5 external dataset references. This vulnerability affects multiple versions of the product across all supported platforms. The issue poses a significant risk due to the potential exposure of sensitive data.
Details
| Severity | Base Score | CVE |
|---|---|---|
| HIGH | 7.1 | CVE-2026-1669 |
Exploitation
To exploit this vulnerability, an attacker can create a malicious .keras model file that includes HDF5 external dataset references. By convincing a user to load this crafted model file, the attacker can trigger the arbitrary file read functionality, allowing them to access and read local files on the victim’s system, potentially leading to the disclosure of sensitive information.
Affected Products
- Keras: Affected versions include 3.0.0 through 3.13.1, which are vulnerable to arbitrary file read due to HDF5 integration.
Mitigation
To mitigate this vulnerability, users should upgrade to the latest version of Keras where the issue has been addressed. Additionally, users should avoid loading model files from untrusted sources to minimize the risk of exploitation. Implementing strict access controls and monitoring for unusual file access patterns can also help in reducing the impact of this vulnerability.