AI Safety Governance Framework 2.0

Summary

Version 2.0 of China's national AI safety governance framework. Building on version 1.0 (2024), its principles include preventing loss of control and keeping AI under human control. It classifies AI safety risks as inherent technical, application and derivative risks, up to catastrophic risk, and sets technical countermeasures, comprehensive governance measures and safety guidelines for development and use. Annexes set principles for grading risks into five levels by application scenario, level of intelligence and scale, the top level having catastrophic and systemic features, and basic principles for trustworthy AI such as ultimate human control.

Safety topics

Status notes

Released on 15 Sept 2025 at the main forum of the 2025 National Cybersecurity Publicity Week. TC260 prepared it under CAC guidance with CNCERT, which led the drafting, and issued it as a TC260 technical document. It also calls for registering AI systems used in critical information infrastructure, for national standards on classifying and grading AI application security risks, for an AI safety evaluation system, and for developers to test regularly whether models may pose loss-of-control risks. Replaced by version 3.0 on 14 Sept 2026 (CN-17).

Sources

Open the official source https://www.cac.gov.cn/2025-09/15/c_1759653448369123.htm

Secondary source: https://www.tc260.org.cn/front/postDetail.html?id=20250915124214

Local copy

Verification

History

1. 2026-10-18related: [{"slug": "cn-ai-safety-governance-framework-1-0", "id": "CN-07"}, {"slug": "cn- → [{"slug": "cn-ai-safety-governance-framework-1-0", "id": "CN-07"}, {"slug": "cn- — two-way link added: CN-12 link(s) to CN-132026-10-18-fix-related-links · Claude Code (claude-opus-5-5)
2. 2026-10-14related: [{"slug": "cn-ai-safety-governance-framework-1-0", "id": "CN-07"}, {"slug": "cn- → [{"slug": "cn-ai-safety-governance-framework-1-0", "id": "CN-07"}, {"slug": "cn- — two-way link to CN-19, CN-20, CN-22 added in run 2026-10-14-discover-leads-2026-10-142026-10-14-discover-leads-2026-10-14 · Claude Code (claude-opus-5-5)
3. 2026-10-14english_version: – → {"kind": "authentic", "url": "https://www.cac.gov.cn/rootimages/uploadimg/175965 — Authentic English text in the issuers' own PDF, a stored source.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
4. 2026-10-14notes: Released at the main forum of the 2025 National Cybersecurity Publicity Week. Dr → Released on 15 Sept 2025 at the main forum of the 2025 National Cybersecurity Pu — Added the governance measures on critical information infrastructure, classification standards, evaluation and loss-of-control testing from the text.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
5. 2026-10-14safety_topics: ["risk-based classification", "evaluation & testing"] → ["risk-based classification", "evaluation & testing", "catastrophic risk", "huma — Added catastrophic risk (section 3, Annex 1), human oversight (principle 1.5, Annex 2) and international cooperation (5.14), all in the text.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
6. 2026-10-14summary: Revised version of the national AI safety governance framework. It tracks changi → Version 2.0 of China's national AI safety governance framework. Building on vers — Rewritten from the Chinese framework text; the earlier summary was under 250 characters and added 'technical' to 防范治理措施.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
7. 2026-10-14language: Chinese → Chinese, English — The PDF carries the issuers' English text after the Chinese.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
8. 2026-10-14original_title: 人工智能安全治理框架（2.0版） → 人工智能安全治理框架2.0 — Taken verbatim from the PDF cover (人工智能 / 安全治理框架2.0); the earlier form with parentheses does not occur in the stored texts.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)
9. 2026-10-14title: AI Safety Governance Framework (Version 2.0) → AI Safety Governance Framework 2.0 — The issuers' own English title on the PDF cover is 'AI Safety Governance Framework 2.0'.2026-10-14-refresh-translation-fixes · Claude Code (claude-opus-5-5)

Related records

AI Safety Governance Framework 1.0 China · Superseded

AI Safety Governance Framework 3.0 China · Published (non-binding, current)

TC260 draft guide on agentic AI system development security China · Draft / proposed

TC260-PG-20268A (AI application security, general provisions) China · Published (non-binding, current)

TC260-PG-202610A (AI application security, health) China · Published (non-binding, current)

AI+ Action Opinions China · Published (non-binding, current)

Citation

National Cybersecurity Standardization Technical Committee (TC260); drafting led by CNCERT under CAC guidance. AI Safety Governance Framework 2.0. 2025-09-15. https://www.cac.gov.cn/2025-09/15/c_1759653448369123.htm. Status: Superseded; verified 2026-10-14 (level A). Local copy archive/CN-13/2026-10-11/official.html retrieved 2026-10-11 14:00:23 UTC, SHA-256 d839994128d59578bd116fe312c29b7ac0c01c2c87c55c046fe97ee675fbe277. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record CN-13.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.