nav emailalert searchbtn searchbox tablepage yinyongbenwen piczone journalimg journalInfo journalinfonormal searchdiv searchzone qikanlogo popupnotification paper paperNew
2026, 02, v.10 65-82
论风险预防原则在人工智能全球治理中的适用
基金项目(Foundation): 中央网信办委托课题“网信前沿新兴技术领域国际合作的新趋势新特点研究”(20252201)的阶段性成果之一
邮箱(Email):
DOI: 10.13871/j.cnki.whuilr.2026.02.005
发布时间: 2026-06-11
出版时间: 2026-06-11
网络发布时间: 2026-06-11
移动端阅读
摘要:

风险预防原则是平衡人工智能发展与安全的重要制度工具,但该原则如何适用于人工智能全球治理仍然存在不确定性。欧洲委员会《人工智能框架公约》确立了风险预防原则的适用方式,其经验具有重要参考价值。在内涵方面,风险预防原则保障的法益范围从生命权、健康权向普遍人权扩张,其构成要素包括风险、科学不确定性、预防措施和证明机制四项内容。在外延方面,风险预防原则正逐渐发展为一项具有义务属性的习惯国际法规则,其实体性义务应包括人工智能测试与审查的注意义务、人类控制与监督义务、透明度义务和救济义务;程序性义务应包括风险准备、风险评估、风险管理和风险通报义务。中国应积极促进风险预防原则适用于人工智能全球治理。具体包括:在国内法层面,完善人工智能相关法律法规中的国际合作条款;在国际法层面,主张风险预防原则涵盖“风险”和“危险”预防,并提出相应的实体性和程序性义务。

Abstract:

The Precautionary Principle serves as a crucial institutional tool for balancing the development and security of Artificial Intelligence(AI). However, uncertainties persist regarding how this principle can be applied to global governance of AI.The European Commission' s Framework Convention on AI establishes the application methods of the Precautionary Principle and provides valuable reference. In terms of conceptual connotation, the scope of legal interests protected by the Precautionary Principle has expanded from the right to life and health to encompassing overall human rights.It also includes four basic elements: risk, scientific uncertainty, precautionary measures, and verification mechanisms. In terms of institutional extension, the Precautionary Principle is evolving into a customary international law rule with obligatory attributes. Its substantive obligations should include duties of care in AI testing and review, human control and supervision, transparency, and relief; procedural obligations should include duties of risk preparation, risk assessment, risk management, and risk communication. China should actively advocate for the application of the Precautionary Principle to global AI governance: at the domestic law level, it should improve the “international cooperation” provisions in AI-related laws and regulations; at the international law level, it should advocate for the Precautionary Principle to cover the prevention of “risks” and “hazards”, and propose corresponding substantive and procedural obligations.

参考文献

[1]黄志雄等:《人工智能的跨国性风险及其国际法治理》,《数字生态与治理》(第3辑),社会科学文献出版社2025年版,第23页。

[2]截至2026年3月20日,已有美国、英国、日本等18国以及欧盟签署该公约,该公约已于2026年1月生效。Council of Europe,The Framework Convention on Artificial Intelligence, https://www.coe.int/en/web/artificial-intelligence/the-framework-convention-on-artificial-intelligence, visited 20 March 2026.

[3]“基于权利”的治理路径旨在确保法律规定能保护个人的权利和自由,强调人权;“基于风险”的治理路径旨在根据对目标风险的评估,对监管行动进行优先排序。UNESCO, Consultation Paper on AI Regulation Emerging Approaches Across the World,https://unesdoc.unesco.org/ark:/48223/pf0000390979, visited 20 March 2026.

[4]沈伟、吴柄萱:《人工智能治理需要什么样的国际条约?——以欧洲委员会〈人工智能与人权、民主和法治框架公约〉为对象》,《德国研究》2025年第2期,第93页。

[1]“Precautionary Principle”通常被译为预防原则或风险预防原则,也有学者译为警惕原则、预防和预警原则、谨慎原则、防备原则、风险防范原则等。苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第35页。

[2]Trail Smelter(United States v. Canada), Reports of International Arbitral Awards, 1941, p.1965.

[3]张华:《论欧盟食品安全法中的风险预防原则:问题与前瞻》,《欧洲研究》2011年第4期,第101页。

[4]欧洲法院和欧洲初审法院在疯牛病案、辉瑞公司案、磷脂质案等案判决中,欧盟委员会在《风险预防原则通讯》等软法性文件中确立了风险预防原则在欧盟法中一般法律原则的地位。例如,欧洲初审法院在磷脂质案中表示,风险预防原则不仅适用于环境领域,而且应该适用于欧盟行动的所有领域,其法律依据是《欧盟条约》中关于公共健康与安全的“目的性条款”;风险预防原则作为欧盟法的一项一般法律原则,要求有关当局采取措施预防对公共健康和安全可能构成的潜在风险。Artegodan and Others v. Commission, the Court of First Instance of the European Communities, Joined Cases T-74/00, T-76/00, T-83/00 to T-85/00, T-132/00, T-137/00 and T-141/00, Judgment of November 2002, para.183.

[5]刘韵清:《预防原则的习惯国际法地位分析》,《国际法研究》2020年第4期,第69页;张华:《论欧盟食品安全法中的风险预防原则:问题与前瞻》,《欧洲研究》2011年第4期,第116页。

[6]马新民:《在中国国际法学会2024年学术年会上的主旨报告》,https://www.fmprc.gov.cn/wjdt_674879/sjxw_674887/202404/t20240430_11291422.shtml,2026年3月20日访问。

[7]Aida Ponce Del Castillo, Artificial Intelligence, Labour and Society 14(ETUI Aisbl 2024).

[8]Tuomas Pöysti, The Precautionary Approach Design Pattern, 5 Digital Society 10(2024).

[9]张涛:《风险预防原则在个人信息保护中的适用与展开》,《现代法学》2023年第5期,第71页。

[10]2002年,欧洲初审法院在辉瑞公司案中首次明确将“风险”“科学不确定性”与风险预防原则联系起来。Pfizer Animal Health SA v. Council of the European Union, the Court of First Instance of the European Communities, Case T-13/99, Judgment of September 2002.有学者认为,“风险”是启动风险预防原则的必要前提,“科学不确定性”是风险概念的固有内涵。因而,“风险”和“科学不确定性”被视为风险预防原则的两项核心要素。苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第39页;金自宁:《科技不确定性与风险预防原则的制度化》,《中外法学》2022年第2期,第504页;张涛:《风险预防原则在个人信息保护中的适用与展开》,《现代法学》2023年第5期,第55页。

[11]郭玉新:《人工智能开源创新的发展型治理》,《法学评论》2026年第2期,第146页。

[12]Aida Ponce Del Castillo, Artificial Intelligence, Labour and Society 14(ETUI Aisbl 2024).

[13]张凌寒:《人工智能法律治理的路径拓展》,《中国社会科学》2025年第1期,第102-103页。

[1]有学者认为,风险预防原则应当成为政府以法律手段应对科技风险的主要原则。陈景辉:《捍卫预防原则:科技风险的法律姿态》,《华东政法大学学报》2018年第1期,第59页。

[2]赵精武:《“元宇宙”安全风险的法律规制路径:从假想式规制到过程风险预防》,《上海大学学报(社会科学版)》2022年第5期,第103页。

[3]唐瑭、王普:《生态环境风险预防原则的法律构造及其功能阐释》,《浙江工商大学学报》2024年第1期,第73页。

[4]唐瑭、王普:《生态环境风险预防原则的法律构造及其功能阐释》,《浙江工商大学学报》2024年第1期,第75页。

[5]ISO, 14971:2019 Medical Devices-Application of Risk Management to Medical Devices, https://www.iso.org/obp/ui/#iso:std:iso:14971:ed-3:v1:en, visited 20 March 2026.

[6][德]尼克拉斯·卢曼:《风险社会学》,孙一洲译,广西人民出版社2020年版,第51页。

[1]《人工智能框架公约》第16条和第20条两次使用“Prevention”,而并未使用“Precaution”或“Precautionary”这一“风险预防”的固定搭配。

[2]张华:《论欧盟食品安全法中的风险预防原则:问题与前瞻》,《欧洲研究》2011年第4期,第105页。

[3]“风险准备”一词常见于金融(证券、会计等)行业,指金融实体为应对可能发生的风险而提前预留资金,以维持自身健康经营并保障客户权益。本文语境中,“风险准备”指为预防人工智能开发、部署和应用活动中可能产生的风险,在进行风险评估前,以“零信任”的态度积极消除可能产生风险的因素。

[4]《人工智能框架公约》第16(2)(g)条、第14(2)(a)条以及第20条。

[5]《人工智能框架公约》第16(2)(e)条、第16(2)(f)条、16(3)条以及第16(4)条。

[6]Noah M. Sachs, Rescuing the Strong Precautionary Principle from Its Critics, 4 University of Illinois Law Review 1295(2011).

[7]苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第37页。

[8]《人工智能框架公约》第24条、第25条。

[1]在国际法和国内法中,风险预防原则存在多种定义和理解。即使在欧盟法中,风险预防原则作为一项一般法律原则也不存在统一界定。沈岿:《风险预防原则与食品添加剂准入裁量——以面粉增白剂去留之争为例》,《北京行政学院学报》2012年第6期,第10页;张华:《论欧盟食品安全法中的风险预防原则:问题与前瞻》,《欧洲研究》2011年第4期,第105页;刘韵清:《预防原则的习惯国际法地位分析》,《国际法研究》2020年第4期,第70页。

[2]有学者认为,风险预防原则的适用前提是生命权、健康权、国家安全等重大法益可能遭受侵害。苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第39页。

[3]从内容上看,人权可分为三种类型:人身人格权利,政治权利以及经济、社会和文化权利。程新宇、杨佳:《人工智能时代人权的伦理风险及其治理路径》,《湖北大学学报(哲学社会科学版)》2024年第3期,第159页;徐明:《个人数字权利实现的立体均衡逻辑》,《法学评论》2025年第4期,第134页。

[4]UNESCO, The Precautionary Principle, https://unesdoc.unesco.org/ark:/48223/pf0000139578, visited 20 March 2026.

[5]发展权也是一项基本人权。张旭:《发展权的基本原则:理论反思与规范表达》,《人权研究》2021年第2期,第355页。

[1]《人工智能框架公约》第5条、第10条。

[2]刘权:《论个人信息保护合规审计——以〈个人信息保护法〉第54、64条为中心》,《法学评论》2026年第1期,第137页。

[3]例如,诺亚·萨克斯(Noah Sachs)认为,风险预防原则至少包括“危害发生的预期(即风险)”“不确定条件”和“预防措施”三项要素。尼尔·A.曼森(Neil A. Manson)从“损害状况(即风险)”“知识状况(即科学不确定性)”和“一般应对措施”三方面强调了风险预防原则的内在构造。托比亚斯·阿诺德森(Tobias Arnoldussen)也认为风险预防原则存在“风险”“不确定性”和“预防措施”三项基本要素。Noah Sachs, Rescuing the Strong Precautionary Principle from its Critics, 4 University of Illinois Law Review 1295(2011); Neil A. Manson, Formulating the Precautionary Principle, 3 Environmental Ethics 265(2002); Tobias Arnoldussen, Precautionary Logic and A Policy of Moderation, 2 Erasmus Law Review 266-271(2009).

[4]苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第39页;谭冰霖:《行政处罚的预防目的及其规范建构》,《法学研究》2024年第6期,第90页;曾庆醒:《涌现效应下的生成式人工智能数据污染及其治理路径》,《法学杂志》2024年第5期,第59页。

[5]苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第40页。

[6]Pfizer Animal Health SA v. Council of the European Union, the Court of First Instance of the European Communities, Case T-13/99, Judgment of September 2002.

[1]苏宇:《风险预防原则的结构化阐释》,《法学研究》2021年第1期,第37页。

[2]人工智能、核能等技术应用,无法实现“零风险”,应容许“剩余风险”存在。刘明全:《论日本核污染的法律规制》,《法学评论》2022年第1期,第153页; Jens Kersten, A Farewell to Residual Risk? A Legal Perspective on the Risks of Nuclear Power after Fukushima, in Jens Kersten&Markus Vogt et al.(eds.), Europe after Fukushima:German Perspectives on the Future of Nuclear Power, https://www.environmentandsociety.org/sites/default/files/1201_fukushima_web_color.pdf, visited 20 March 2026.

[3]胡东海:《“谁主张谁举证”规则的法律适用》,《法学》2019年第3期,第92页。

[4]James Cameron&Juli Abouchar, The Precautionary Principle:A Fundamental Principle of Law and Policy for the Protection of the Global Environment, 1 Boston College International and Comparative Law Review 12(1991).

[5]例如,在欧共体转基因农产品案中,欧盟主张风险预防原则构成国际法的一般法律原则或习惯国际法规则;美国反对风险预防构成一项国际法规则,因其缺乏一致表达,概念存在不确定性。European Communities, Measures Affecting the Approval and Marketing of Biotech Products, Panel Report on European Communities, WT/DS291/R, WT/DS292/R, 29 September 2006.

[1]UNESCO, Recommendation on the Ethics of Artificial Intelligence, SHS/BIO/REC-AIETHICS/2021, 24 November 2021, para.12.

[2]UN, GA Resolution, Seizing the Opportunities of Safe, Secure and Trustworthy Artificial Intelligence Systems for Sustainable Development, A/78/L.49, 11 March 2024, p.5.

[3]Bin cheng, Studies in International Space Law 125(Oxford University Press 1997).

[1]ILC, Draft Articles on Prevention of Transboundary Harm from Hazardous Activities, with Commentaries, 2001, p.148.

[2]多数学者认为,国际法委员会二读通过的《预防危险活动的跨境损害条款草案》属于习惯国际法的编纂。胡绪雨:《跨境污染损害责任的私法化》,《政法论坛》2013年第6期,第49页;罗旷怡:《国际法必要勤勉概念的演变与澄清》,《南大法学》2025年第6期,第145页。

[3]“国际法不加禁止的危险活动造成跨境损害”包含四个要素:该活动未被国际法禁止、具有引起重大损害的风险、损害具有跨界性以及损害造成“有形后果”。邵莉莉:《论国际生物安全损害法律救济模式的完善》,《当代法学》2023年第6期,第149页。

[4]金自宁:《科技不确定性与风险预防原则的制度化》,《中外法学》2022年第2期,第504页。

[5]ILC, Draft Articles on Prevention of Transboundary Harm from Hazardous Activities, with Commentaries, 2001, p.155.

[6]Pulp Mills on the River Uruguay(Argentina v. Uruguay), Judgment, ICJ Reports, 2010, para.71.

[1]《人工智能框架公约》第13条。

[2]《负责任的生成式人工智能建议书》第5条、第6条、第13条、第14条、第15条。WEF, The Presidio Recommendations on Responsible Generative Al, https://www3.weforum.org/docs/WEF_Presidio_Recommendations_on_Responsible_Generative_AI_2023.pdf, visited 20 March 2026.

[3]截至2026年3月20日,已有58国支持并联合实施该宣言。U.S. Department of State, Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy, https://www.state.gov/political-declaration-on-responsible-military-use-of-artificial-intelligence-and-autonomy/, visited 20 March 2026.

[4]王泽鉴:《侵权行为》,北京大学出版社2009年版,第242页。

[5]European Commission, Speech by President von der Leyen at the 15th Congress of the European Trade Union Confederation,https://ec.europa.eu/commission/presscorner/detail/en/speech_23_2926, visited 20 March 2026.

[6]《全球数字契约》第55(d)条。UN, GA Resolution, The Pact for the Future(and Its Annexes), A/79/L.2, 2024.

[7]《人工智能可持续发展决议》第6(k)条。UN, GA Resolution, Seizing the Opportunities of Safe, Secure and Trustworthy Artificial Intelligence Systems for Sustainable Development, A/78/L.49, 2024.

[8]UN, GA Resolution, Artificial Intelligence in the Military Domain and Its Implications for International Peace and Security, A/C.1/79/L.43, 2024.

[9]Aida Ponce Del Castillo, Artificial Intelligence, Labour and Society 19(ETUI Aisbl 2024).

[1]张磊:《人工智能国际规则博弈态势、共识分歧与中国“话语势能”建构》,《新疆社会科学》2024年第4期,第127页。

[2]《全球数字契约》第55(d)条。UN, GA Resolution, The Pact for the Future(and Its Annexes), A/79/L.2, 2024.

[3]《人工智能伦理问题建议书》第11条。UNESCO, Recommendation on the Ethics of Artificial Intelligence, SHS/BIO/REC-AIETHICS/2021, 24 November 2021.

[4]刘文杰:《何以透明,以何透明:人工智能法透明度规则之构建》,《比较法研究》2024年第2期,第120页。

[5]《人工智能框架公约》并未直接为人工智能企业等私主体设定义务,缔约国通过国内立法转化或批准直接适用等程序后,企业才需遵守相关规则。

[6]《人工智能框架公约》第14(2)(a)条、第15(2)条。

[7]《人工智能框架公约》第19条、第23(2)条、第24条、第25条。

[8]刘文杰:《何以透明,以何透明:人工智能法透明度规则之构建》,《比较法研究》2024年第2期,第125-129页。

[1]OECD, Towards a Common Reporting Framework for AI Incidents, https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/02/towards-a-common-reporting-framework-for-ai-incidents_8c488fdb/f326d4ac-en.pdf, visited 20 March 2026.

[2]Marietjie Botes, Regulating Scientific and Technological Uncertainty:The Precautionary Principle in the Context of Human Genomics and AI, 5 South African Journal of Science 3(2023).

[3]《人工智能伦理问题建议书》第12条。UNESCO, Recommendation on the Ethics of Artificial Intelligence, SHS/BIO/REC-AIETHICS/2021, 24 November 2021.

[4]《人工智能框架公约》第14(2)(b)条、第14(2)(c)条。

[5]Bronwyn Howell, Regulating Artificial Intelligence in a World of Uncertainty, https://www.aei.org/research-products/report/regulating-artificial-intelligence-in-a-world-of-uncertainty/, visited 20 March 2026.

[6]王金鹏:《跨界环境损害预防中的国家勤勉尽责义务刍议》,《甘肃政法大学学报》2022年第2期,第124页。

[7]张华:《论欧盟食品安全法中的风险预防原则:问题与前瞻》,《欧洲研究》2011年第4期,第105页。

[1]《负责任军事使用人工智能与自主技术的政治宣言》第3条、第4条、第9条。U.S. Department of State, Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy, https://www.state.gov/political-declaration-on-responsible-military-use-of-artificial-intelligence-and-autonomy/, visited 20 March 2026.

[2]《先进人工智能系统组织的广岛进程国际行为准则》第6条、第7条、第11条。EU, Hiroshima Process International Code of Conduct for Organizations Developing Advanced Al Systems, https://digital-strategy.ec.europa.eu/en/library/hiroshima-process-international-code-conduct-advanced-ai-systems, visited 20 March 2026.

[3]European Commission, Regulation No.178/2002, Art.3(10).

[4]European Commission, Regulation No.178/2002, Art.3(11).

[5]《人工智能框架公约》第14(2)(a)条。

[6]European Commission, Regulation No.178/2002, Art.3(11).

[1]European Commission, Regulation No.178/2002, Art.3(13).

[2]《中国关于加强人工智能伦理治理的立场文件》《全球人工智能治理倡议》《新一代人工智能伦理规范》《人工智能安全治理框架2.0》《人工智能全球治理行动计划》等。

[3]2025年中国全国网络安全标准化技术委员会《人工智能安全治理框架2.0》第3章。

[4]2025年中国国家互联网信息办公室《全球人工智能治理倡议》第5条。

[5]《国家互联网信息办公室关于〈人工智能拟人化互动服务管理暂行办法(征求意见稿)〉公开征求意见的通知》,https://www.cac.gov.cn/2025-12/27/c_1768571207311996.htm,2026年3月20日访问。

[1]《中华人民共和国国民经济和社会发展第十五个五年规划纲要》,https://www.gov.cn/yaowen/liebiao/202603/content_7062633.htm,2026年3月20日访问。

[1]唐瑭、王普:《生态环境风险预防原则的法律构造及其功能阐释》,《浙江工商大学学报》2024年第1期,第74页。

基本信息:

DOI:10.13871/j.cnki.whuilr.2026.02.005

中图分类号:D99;TP18

引用信息:

[1]张磊.论风险预防原则在人工智能全球治理中的适用[J].武大国际法评论,2026,10(02):65-82.DOI:10.13871/j.cnki.whuilr.2026.02.005.

基金信息:

中央网信办委托课题“网信前沿新兴技术领域国际合作的新趋势新特点研究”(20252201)的阶段性成果之一

发布时间:

2026-06-11

出版时间:

2026-06-11

网络发布时间:

2026-06-11

检 索 高级检索

引用

GB/T 7714-2015 格式引文
MLA格式引文
APA格式引文