Skip to content

GRF · 风险治理

数字与人工智能治理

各国政府和运营机构依靠人工智能和数字公共 系统 来提供服务、分配资源和发现风险。规则正在变化:欧盟《人工智能法案》(2024)自 2025 年起分阶段实施,欧洲委员会《人工智能公约》于 2024 年开放签署,2025 年联合国设立了人工智能科学小组和人工智能治理全球对话。全球风险论坛(Global Risks Forum,GRF)召集主管机构、技术专家和权利领域专家,探讨如何保持监督、问责和权利不受削弱。GRF 不批准任何系统。

02治理问题

机构需要回答的三个问题。

监督

人工监督

Who can understand, question and override an automated decision, and do they have the authority and time to do so?

问责

可问责性

Who answers for an automated decision that harms someone, and how can the affected person seek review?

权利

权利与包容

How are privacy, non-discrimination and access protected, including for people who cannot use digital channels?

03The instruments

A layered and still moving rulebook.

工具机构功能状态说明
《人工智能法案》欧洲联盟Risk-tiered obligations: prohibited practices, duties for general-purpose models, then high-risk system requirements分阶段适用;2026 年的《数字综合方案》(Digital Omnibus)协议调整了高风险条款的时间表,因此日期应以《欧盟官方公报》为准
《人工智能与人权、民主和法治框架公约》Council of Europe, open to non-members对公共当局使用、也可由私营主体自愿采用的 AI 系统 ,在其整个生命周期内具有约束力的义务2024 年 9 月开放签署;批准情况请查阅条约办公室
OECD 人工智能原则经合组织(OECD),许多非成员亦予遵守Values-based principles and recommendations to governments, updated in 2024In force as a Recommendation
《全球数字契约》联合国大会,《未来契约》附件Commitments on digital inclusion, data governance and AI, including an independent scientific panel and a global dialogue专家小组已提交报告,对话首次会议于 2026 年 7 月在日内瓦举行
《人工智能伦理问题建议书》UNESCOReadiness assessments and ethical impact assessment for member statesImplementation through national readiness assessments
04公共部门使用

The governance question for a ministry deploying a model.

当机构使用算法确定检查优先次序、分配福利或对紧急呼叫分级时,首先是制度问题,然后才是技术问题。谁批准了这一使用?谁能叫停?受影响的人如何对结果提出异议?哪个监管机构负责审计?GRF 对话让部署机构、其监督机构、数据保护机构和受影响用户围绕同一案例展开讨论,并记录他们一致认可的监督链条。

Technical assurance methods sit with GCRI’s AI and autonomous systems platform; model risk in finance with GRA’s AI and financial model risk platform; the specialist community is AI and Digital Governance.

05International guidance

构建对话框架的工具。

  • 《联合国教科文组织人工智能伦理问题建议书》(2021年)。
  • The OECD AI Principles (2019, updated 2024).
  • The Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (2024), for the states that are party to it.
  • The Global Digital Compact (2024) and international human rights law, including the rights to privacy and to an effective remedy.
  • 适用于各参与机构的国家和区域法律,例如欧盟《人工智能法》。

Named instruments are references for dialogue, not claims of alignment. GRF does not assess any institution’s conformity with them, and it does not regulate, finance, procure or certify.

06Exercising oversight

Testing what happens when an automated system is wrong.

Oversight never exercised is an assumption. In Universe, a scenario can have an automated triage, warning or eligibility system produce a confident but wrong output. The exercise asks: do the people responsible notice, can they override, can they explain the decision afterwards? Evaluators record where oversight held and where it failed. The exercised institutions review findings before they inform Reports. A missing common practice can become a change proposal to the Nexus Standards.

  1. 01

    监督对话

    Structured dialogues that produce a record of oversight questions, responsibilities and open issues for a given public system.

  2. 02

    专业社群

    Innovation Nexus works on governed public-interest innovation, and Governance Nexus on comparative governance under technology change.

  3. 03

    专业平台

    人工智能与数字治理 是该领域的专业社区,设有数字权利与数据治理章节。

07对口方与联系人

三个机构中的同一领域。

全球风险与创新中心(Global Centre for Risk and Innovation,GCRI)研究 人工智能 和负责任技术。全球风险联盟(Global Risks Alliance,GRA)研究金融领域的 人工智能与模型风险 。全球风险论坛(Global Risks Forum,GRF)的相关领域: 信息与公众信任 以及决策权和责任。若要将某一系统及其监督问题提交对话,请通过 联系表单提交;确认回复会提供参考编号并说明下一步。

09问题

问题与解答。

GRF 是否审计人工智能系统?
不会。GRF 召集并记录关于监督与问责的对话。它不审计、评估、认证或批准任何系统。
演练可以使用在线运行的 AI 系统吗?
Only within an agreed, contained test environment. Exercises never act on real people’s cases, and real infrastructure is never an unapproved target.
公共机构如何将一个系统带入对话?
Write to the Secretariat at the contact form with the system and the oversight question. The acknowledgement gives a reference and names the next step.
报告贡献

从本页贡献证据。

Reports 是联合体的季刊。您对该主题的了解可以成为下一期中的证据、案例或更正,署名归您,并在属于您本人作品时以 CC BY 4.0 发布。

全球风险论坛(Global Risks Forum)
全球风险论坛(Global Risks Forum)

把一个治理问题带到论坛。

信息和咨询现已开放。每项咨询都会获得一个编号,并告知下一步;申请、付款、算力分配和注册只有在相应团队、条款和审查能力就绪时才会开放。