CAC, TC260, NFRA, PBOC, CSRC
Mainland China
Mainland China AI regulatory stack - CAC measures, TC260 standards, NFRA / PBOC / CSRC financial rules ยท Binding administrative measures and national standards; National AI Law in drafting
Not one statute but a vertical stack. CAC-led measures cover recommendation algorithms (2022), deep synthesis (2023), generative AI services (Interim Measures, 2023) and labelling of AI-generated content (2025, with mandatory standard GB 45438-2025). TC260 turns these into engineering standards - the AI Safety Governance Framework 3.0 and GB/T 45654, 45652 and 45674 of 2025 - the operational equivalent of ISO/IEC 42001 and NIST AI RMF. In finance, the NFRA's 32 measures, PBOC JR/T standards and CSRC programme-trading rules apply. The CSL, DSL and PIPL underpin all of it. A draft National AI Law proposes basic, significant and high-risk tiers.
Who it binds. Anyone offering public-facing AI services within Mainland China, including offshore providers; banks, insurers and securities firms under NFRA, PBOC and CSRC; and anyone moving Chinese data across the border.
Start here
The questions decision-makers ask most about Mainland China. Each answer cites the provision and the SHIELD dimension it belongs to.
Ask about Mainland China
Answers are drawn from the documentation - for orientation, not legal advice.
Your question is sent to our AI provider to generate an answer. Please do not include personal or confidential information.
Compare with the others
EU AI Act
Hard law that sorts AI uses into risk tiers and attaches duties to each tier.
ISO/IEC 42001
A certifiable management system for AI - effectively the auditor's checklist.
SFC / HKMA guidance
Market-specific expectations that make named senior individuals personally accountable.
NIST AI RMF 1.0
A voluntary, plain-language structure: Govern, Map, Measure, Manage.
MAS AI risk management
Singapore financial-sector expectations for AI oversight, risk assessment and lifecycle controls.
Want to explore your organisation's governance maturity against the 5 assessed frameworks? Take the SHIELD self-assessment.