模型供應鏈來源不透明

Risk Description

AI models used by organizations may pass through multiple hands (foundation model provider → fine-tuning vendor → system integrator → final deployer), with each stage potentially introducing unknown modifications. Without a complete model provenance chain, organizations cannot confirm whether the model in use matches the claimed version, nor assess whether intermediate stages introduced security risks or biases.