The advancement of artificial intelligence (AI) technology has posed significant challenges to traditional copyright systems, particularly with the emergence of generative AI, which has disrupted conventional copyright industries. Balancing copyright protection with AI innovation necessitates expertise in legislative, judicial, and enforcement domains. The rapid proliferation of large-scale models and their applications has intensified debates over whether and how generative AI service providers should fulfill copyright obligations. This paper adopts the position that such providers must exercise appropriate copyright due diligence, aiming to refine the subjects, scope, and legal consequences of these obligations. This research employs a multidisciplinary methodology, integrating historical analysis to trace the evolution of AI-related copyright norms; literature review to synthesize global regulatory frameworks and scholarly discourse; normative analysis of legal provisions, including China's Interim Measures for Generative AI Management; comparative studies of international legislative practices; and case studies of judicial rulings on AI-generated content disputes. By integrating existing legal frameworks with empirical insights, this paper proposes a copyright due diligence framework for generative AI service providers. Specifically, generative AI service providers should implement primary responsibility for copyright governance and fulfill reasonable copyright duty of care. This duty does not equate to ex-ante filtering obligations; service providers should qualify for "safe harbor" protections but must avoid adopting an "ostrich policy"—taking necessary measures to prevent infringement and mitigate damages after receiving valid infringement notices. The establishment of such duty should avoid overly stringent standards, instead considering technological limitations and practical constraints. It should align with the principle of utmost good faith and the diligence obligations of a prudent manager, ensuring flexibility and adaptability. Providers must adopt comprehensive copyright management mechanisms, including: (1) preventive measures: Ensuring model explainability, disclosing training data, and establishing content labeling systems; (2) real-time monitoring: Developing robust user guidelines and complaint-handling mechanisms; and (3) post-infringement actions: Implementing user relationship governance and bearing adverse consequences of burden of proof reversal in specific cases. Generative AI service providers bear fault-based liability for copyright infringement. If infringement occurs despite compliance with duty-of-care standards, liability should be exempted. This study clarifies critical ambiguities in AI copyright governance by rejecting the “ostrich policy” of willful negligence while affirming the “safe harbor” principle; providing standardized criteria for courts to adjudicate infringement claims; and offering a compliance roadmap for AI developers to foster innovation without compromising creator rights. This paper advocates transcending the micro-level "liability-centric" framework and instead constructing the copyright protection obligations of generative AI service providers through a macro-level governance perspective, tailored to the constraints of existing technological conditions. Specifically, it argues that traditional approaches overemphasize post-infringement liability allocation while neglecting systemic governance mechanisms to prevent copyright risks at their source. This paper argues that the copyright governance framework for generative AI service providers should adopt a comprehensive lifecycle approach—spanning pre-emptive, real-time, and post-infringement measures—while ensuring that the imposition of copyright duty of care aligns with principles of legitimacy, rationality, and practicality. These contributions advance theoretical discourse on AI law and promote sustainable development across creative and technological sectors.