Xinference (< 2.12.0)

Xinference versions prior to 2.12.0 are vulnerable to remote code execution because they unconditionally enable 'trust_remote_code=True' when loading models, allowing attackers to execute arbitrary Python code via crafted model configurations.