Unsloth Studio Vulnerability Enables Code Execution via Model Inspection

• •Unsloth Studio's flaw allows arbitrary code execution during model inspection.
• •The vulnerability affects users with permissions in enterprise AI environments.
• •Pillar Security reported the flaw, which Unsloth has since patched.

A critical vulnerability in Unsloth Studio, an open-source library for fine-tuning LLMs, allows arbitrary code execution through model inspection. This flaw arises from the 'trust_remote_code=True' setting, which permits execution of Python code from a model's Hugging Face repository upon merely inspecting its metadata. The code runs with the user's permissions, potentially exposing sensitive data, model artifacts, and credentials in enterprise environments. While Unsloth has patched the vulnerability, there are disputes regarding the adequacy of security measures, with Pillar Security emphasizing the risks of untrusted code execution. No evidence of real-world exploitation has been reported, but the potential for significant impact remains. The vulnerability is associated with CVE-2026-46432, published on June 9, 2026.