Unsloth Studio Flaw Turns Routine Model Inspection Into Code Execution

Unsloth Studio fixed a vulnerability where its use of transformers' trust_remote_code=True caused model repository metadata inspection to execute arbitrary Python code, potentially exposing proprietary training data, model artifacts, and credentials; Pillar Security reported the issue and Unsloth released a patch (2026.6.9) though Unsloth disputed parts of the assessment and no CVE was assigned or evidence of in-the-wild exploitation has been observed.