Remote Code Execution in Axolotl via Hugging Face Model

Executive Summary

CVE-2026-86169 is a remote code execution vulnerability in Axolotl versions up to 0.18.0. It occurs because the multipack patch path sets trust_remote_code to None by default, bypassing security checks. Attackers can exploit this by uploading a malicious Hugging Face model repository as base_model, which is loaded with trust_remote_code=True during model initialization, enabling arbitrary Python code execution.

Useful 0 Not Useful 0

Detection Guidance

To detect this vulnerability, check if Axolotl version 0.18.0 or earlier is installed by running: pip show axolotl. If installed, verify if multipack patching is enabled in your configuration files. Look for models loaded with trust_remote_code=True in logs or model loading processes.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability allows attackers to execute arbitrary code on systems using vulnerable Axolotl versions. If you use Axolotl for fine-tuning language models, an attacker could compromise your system, steal data, or disrupt operations by uploading a malicious model. The impact includes potential data breaches, system compromise, and unauthorized access.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability could lead to non-compliance with GDPR and HIPAA due to potential data breaches. GDPR requires protection of personal data, and HIPAA mandates safeguards for protected health information. A remote code execution flaw increases the risk of unauthorized data exposure, resulting in legal penalties, fines, and reputational damage for organizations handling sensitive data.

Useful 0 Not Useful 0

Mitigation Strategies

Immediately upgrade Axolotl to the latest patched version using pip install --upgrade axolotl. Review and update all model loading configurations to explicitly set trust_remote_code=False where applicable. Monitor Hugging Face model repositories for unauthorized or suspicious base_model references.

Useful 0 Not Useful 0