从风险识别到责任修复_AI治理的全球标准路径_61页_2mb

OECD Due Diligence Guidance for Responsible AI Summary

Core Content

The OECD Due Diligence Guidance for Responsible AI is a comprehensive framework designed to help enterprises implement the OECD Guidelines for Multinational Enterprises (MNE Guidelines) and the OECD Recommendation on Artificial Intelligence (AI Principles). It outlines a structured approach to identifying, addressing, and mitigating risks associated with the development and use of AI systems across the entire value chain.

This guidance is intended for multinational enterprises that are involved in the AI system value chain, including:

• Suppliers of AI inputs (e.g., data providers, compute providers, cloud services)
• Developers of AI systems (e.g., those involved in planning, design, model building, deployment, and operation)
• Users of AI systems (e.g., financial institutions, real economy enterprises)

The framework is based on the OECD RBC due diligence approach, which includes six key steps:

1. Embed RBC into policies and management systems
2. Identify and assess actual and potential adverse impacts
3. Cease, prevent, and mitigate adverse impacts
4. Track implementation and results of due diligence activities
5. Communicate actions to address impacts
6. Provide for or cooperate in remediation when appropriate

The guidance emphasizes a risk-based approach to responsible AI, ensuring that enterprises take proactive steps to prevent harm while fostering innovation and trust. It also promotes policy coherence and interoperability between the MNE Guidelines, AI Principles, and other national or international AI risk management frameworks.

Main Viewpoints

• Responsible AI is not just a technical issue, but a broader ethical and social responsibility that involves all stakeholders.
• AI systems can significantly benefit society, but they also pose risks that must be proactively managed.
• Stakeholder engagement and transparency are central to the due diligence process, especially for workers and affected communities.
• The MNE Guidelines and AI Principles are complementary, and this guidance bridges the gap between them by offering practical implementation steps.
• SMEs face unique challenges in implementing due diligence due to limited resources and capacity, and the guidance encourages collaborative approaches and industry initiatives to support them.

Key Information

Scope of the Guidance

• Targeted at enterprises across the AI system value chain, from suppliers to users.
• Not limited to the technology sector, as AI is increasingly used in non-technological contexts.
• Excludes hardware supply chains (e.g., mining and manufacturing of components) which are covered in separate guidance.

Structure and Implementation

• Six steps guide the due diligence process, each with practical examples and a roadmap of related provisions from existing frameworks.
• The guidance is not a checklist, but a tool for adaptation, recognizing that different situations may require different approaches.
• Practical examples are drawn from various AI risk management frameworks and industry consultations.

Special Considerations

• SMEs are encouraged to use collaborative strategies and engage in industry initiatives to reduce costs and improve compliance.
• Stakeholder engagement is emphasized, particularly with workers and affected communities.
• Transparency, explainability, and traceability are critical in the AI system lifecycle.
• Remediation and cooperation are important when adverse impacts occur or are suspected.
• Interoperability and global cooperation are key to aligning with international standards.

Legal and Regulatory Context

• The guidance draws on existing international and national AI frameworks, regulations, and initiatives.
• It supports National Contact Points (NCPs) in promoting responsible AI standards and addressing violations.
• It is also relevant for civil society, workers, trade unions, and regulatory authorities involved in AI oversight.

Strategic Importance

• Responsible AI is increasingly a business requirement, as more enterprises integrate AI risk management into their procurement and operations.
• Demonstrating commitment to responsible AI can lead to competitive advantage, access to capital markets, and trust among investors and regulators.
• The guidance helps navigate complex regulatory environments, ensuring that enterprises can operate effectively across multiple jurisdictions.

Conclusion

The OECD Due Diligence Guidance for Responsible AI is a practical and adaptable tool for enterprises seeking to align with international standards on AI ethics and governance. It supports a holistic, risk-based approach to AI development and use, promoting trust, innovation, and sustainable growth while addressing ethical, social, and economic risks. By following this guidance, enterprises can better meet the expectations of stakeholders, regulators, and investors, ensuring that AI technologies contribute positively to society.