AI Safety Governance Framework v1.0 (National Technical Committee 260 on Cybersecurity of SAC)

• Authority: Other authorities
• Collections: Chinese law and policy;Editors' Picks
• Most recent activity: Enacted
• Most recent activity date: 2024-09-01

Risk subdomain coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the risk subdomain are explicitly described.

• Unfair discrimination and misrepresentation
• Exposure to toxic content
• Compromise of privacy
• AI system security vulnerabilities
• False or misleading information
• Pollution of information ecosystem
• Disinformation, surveillance, and influence at scale
• Cyberattacks, weapon development or use
• Fraud, scams, and targeted manipulation
• Governance failure
• AI pursuing its own goals
• AI possessing dangerous capabilities
• Lack of capability or robustness
• Lack of transparency or interpretability

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the risk subdomain are described.

• Unequal performance across groups
• Overreliance and unsafe use
• Loss of human agency and autonomy
• Power centralization and unfair distribution
• Increased inequality and decline in employment
• Competitive dynamics
• Multi-agent risks

All other risk subdomains had no coverage in this document.

Risk Coverage Summary

The document has comprehensive coverage of approximately 18-20 subdomains across all risk categories. It demonstrates particularly strong focus on AI system safety failures (7.1, 7.2, 7.3, 7.4), malicious actors (4.1, 4.2, 4.3), misinformation (3.1, 3.2), privacy and security (2.1, 2.2), discrimination and toxicity (1.1, 1.2, 1.3), and socioeconomic risks (6.1, 6.2, 6.4, 6.5). The framework provides detailed risk classifications and mitigation strategies across the entire AI lifecycle.

Sector coverage

Good coverage — Comprehensive governance measures or mitigations with clear procedures relating to the sector are explicitly described.

• Information
• Health Care & Social Assistance
• Public Administration (excl. National Security)
• National Security

Minimal coverage — Brief mention but no specific mitigations, recommendations or governance measures relating to the sector are described.

• Trade, Transportation & Utilities
• Finance & Insurance
• Scientific Research & Development Services
• Educational Services

All other sectors had no coverage in this document.

Sectors Summary

The document governs AI use across multiple critical sectors with explicit coverage of government/public administration, healthcare, finance, telecommunications, transportation, and energy/utilities. It provides sector-specific guidance for high-risk applications while establishing general governance principles applicable across all sectors.

Legislative Status

Soft Law

Legislative Status Summary

This is a voluntary governance framework that uses predominantly recommendatory language ('should') and lacks binding enforcement mechanisms or legal penalties. It is presented as a set of principles and guidelines to promote consensus and cooperation among stakeholders rather than legally enforceable obligations.

Actors

Proposer

• Entity Types: AI_Governance_Actor
• Entities: National Technical Committee 260 on Cybersecurity of SAC (Standardization Administration of China)
• Summary: The document is authored by the National Technical Committee 260 on Cybersecurity of SAC, which is a technical standards committee under China's Standardization Administration. The framework is presented as implementing the Global AI Governance Initiative.

Target

• Entity Types: AI_Developer, AI_Deployer, AI_User
• Entities: model and algorithm researchers and developers; service providers; users; technology R&D institutions; government agencies; enterprises; public service units
• Summary: The document explicitly targets multiple categories of actors throughout the AI lifecycle. Section 1.3 clearly identifies the target entities, and Sections 6.1-6.4 provide specific safety guidelines for different target groups including developers, service providers, users in key areas, and general users.

Enforcer

• Entity Types: AI_Governance_Actor
• Entities: government authorities; competent authorities; industry associations
• Summary: The document references government oversight and competent authorities in multiple sections, though enforcement is primarily through voluntary compliance and industry self-regulation rather than formal legal enforcement. Section 5.9 emphasizes industry self-regulation mechanisms.

Monitor

• Entity Types: AI_Governance_Actor, AI_Developer, AI_Deployer
• Entities: government agencies; industry associations; social organizations; service providers; users
• Summary: The document establishes a multi-stakeholder monitoring approach involving government agencies, industry bodies, and the entities themselves. Section 5.7 describes information sharing and emergency response mechanisms, while Section 5.9 establishes social supervision mechanisms.

AI Lifecycle Stages

Lifecycle stages covered by the documents, as defined in the OECD Framework for the Classification of AI Systems

• Plan and Design
• Collect and Process Data
• Build and Use Model
• Verify and Validate
• Deploy
• Operate and Monitor

AI Lifecycle Summary

The document comprehensively covers all stages of the AI lifecycle from planning through operational monitoring. It explicitly addresses design and planning (Section 6.1a), data collection and processing (Sections 3.1.2, 4.1.2), model building and training (Sections 3.1.1, 4.1.1), verification and validation (Sections 6.1g-j), deployment (Section 5.1), and ongoing monitoring and maintenance (Sections 5.7, 6.2f). The framework emphasizes whole-process governance throughout the entire AI lifecycle.

Technical Scope

Terms relating to technical scope that are mentioned or implicitly referenced in the document

• AI Models
• Foundation Models
• AI Systems
• Compute Threshold
• Generative AI
• Open Weight / Open Source

Technical Scope Summary

The document explicitly mentions AI models, AI systems, and AI algorithms throughout. It does not use the specific terms 'frontier AI', 'general purpose AI', 'foundation models', or 'generative AI' as defined categories, though it does reference generative AI capabilities in the context of risks. It mentions compute thresholds for registration requirements and addresses model reuse. The document does not explicitly distinguish between open-weight and closed models.