西安交通大学学报（社会科学版）2025，Vol.45Issue(2)：99-108,10.DOI:10.15896/j.xjtuskxb.202502009
人工智能负面清单管理模式的功能定位和实现路径
Functional Positioning and Implementation Path of the Negative List Management Model for Artificial Intelligence
摘要
Abstract
The rapid iteration of artificial intelligence technology has led to the continuous emergence of new applications and business models,injecting strong momentum into economic and social development.However,the legal challenges it brings are increasingly prominent,mainly manifested in the conflicts between risk uncertainty and legal predictability,the tension between diversification of application scenarios and institutional uniformity,and the contradiction between rapid technological iteration and relatively lagging legal frameworks.The complexity and diversity of AI technology make risk prevention and control highly challenging,with differences in technical logic,application scenarios,and potential risks making it difficult to establish universal regulatory standards.Due to the inherent"black box"characteristics of AI systems,their decision-making processes are difficult to explain through simple linear logic,increasing the difficulty of risk identification and assessment.With the development of large models,the universality and multi-functionality of AI systems continue to emerge,and security risks can be transmitted to the entire society through downstream applications.Even system designers and developers find it difficult to fully anticipate their performance in various situations. Countries worldwide face the dual challenge of balancing industrial development with risk prevention:The EU's AI Act adopts a four-tier risk classification system,but its complex regulatory architecture not only increases implementation difficulties but may also force enterprises to divert significant resources from technological innovation to compliance.The EU AI Office,with only 140 staff members,struggles to manage AI regulation across the entire EU region.The United States,emphasizing market leadership and industry self-regulation,guides development through voluntary risk management frameworks and industry best practices,but this regulatory model may lead to insufficient risk prevention,rapid risk diffusion,and irreparable damage. As a market access control measure,the negative list management model demonstrates three major functional advantages:first,by clearly listing areas requiring strict regulation or restrictions,it provides reasonable expectations for industry development and guides enterprises to pay more attention to compliance in innovation;Second,it abandons the mindset of comprehensively predicting and controlling all risks,focusing instead on controlling major risk directions in AI technology,such as nuclear facilities,biology,cybersecurity,and high-risk technologies like deep fakes;Third,through establishing dynamic adjustment mechanisms,it achieves agile governance of technological iterations,allowing regulatory authorities to continuously optimize list content based on multi-stakeholder participation.Compared to EU and US AI governance approaches,the negative list management model has unique advantages:The"two-tier"risk prevention and control architecture avoids the excessive complexity of the EU scheme,reducing implementation difficulties;the"non-prohibited entry"mechanism effectively reduces enterprise compliance costs,avoiding over-compliance caused by adopting the strictest standards;this model better suits China's actual needs as an AI industry pursuer,reserving policy space for technological breakthroughs while ensuring safety. To ensure the effective operation of the negative list management model in the AI field,three specific implementation paths need to be established:first,establishing procedures for"establishing,amending,and abolishing,"providing behavioral guidance through list content establishment,achieving dynamic adjustment through regular and temporary assessments,and timely abolishing specific items when risks are effectively controlled;Second,clarifying the setting approach for items within the list from three dimensions:specifying the specific direction of risk management(such as AI products,services,or systems),prioritizing application scenarios generally considered to have major risks(such as cybersecurity,chemical,biological,nuclear facilities,etc.),and gradually incorporating industry best practices;Third,constructing a differentiated regulatory system,implementing licensing management for items within the list,focusing on evaluating subject qualifications,technical applications,and security assurance capabilities,while adopting filing management for applications outside the list that have certain social influence,while preventing filing from becoming de facto licensing.Through the organic connection and implementation of the above paths,the healthy and orderly development of the AI industry can be promoted.关键词
人工智能治理/负面清单/人工智能立法/风险规制/市场准入Key words
artificial intelligence governance/negative list/artificial intelligence legislation/risk regulation/market access分类
计算机与自动化引用本文复制引用
张心宇..人工智能负面清单管理模式的功能定位和实现路径[J].西安交通大学学报（社会科学版）,2025,45(2):99-108,10.基金项目
国家社会科学基金重大项目(21&ZD195) （21&ZD195）
国家社会科学基金重点项目(21AFX004). （21AFX004）
