The utilization of works by generative artificial intelligence(AI)has two distinctive features:massive scale and algorithmic processes.These features present substantial challenges in analyzing and assigning copyright infringement liability.The safe harbor rules,based on meeting certain conditions,offer immunity that can address the structural shift to machine training,align with risk allocation goals,and have preventive effects.Given the technical capabilities of service providers,the regulatory needs of large AI models,and the benefits sought by copyright holders,a"two-tier"safe harbor rule for service providers is essential.During the training stage,safe harbor would allow service providers to use works without prior permission,provided they satisfy the obligation of disclosure,identify and respect the reserved rights,obtain indirect economic benefits and provide comprehensive compensation.During the output stage,their obligations should include establishing a complaint mechanism,eliminating redundant data of works,optimizing model performance to prevent overfitting,mitigating malicious user conducts,and implementing request-based copyright filtering.
generative AIcopyright infringementsafe harbor rulesduty of care
万方数据
