AI Safety Index

How does it influence Chinese companies’ behavior?

It is challenging to provide a fair comparison between frontier AI companies in China and those in the United States because of differing contexts. It is not obvious whether companies are more likely to abide by their own voluntary commitments (which are common in the U.S.) or draft laws and government standards that have not yet come into force (which are common in China). To enable our reviewers to draw their own conclusions, we will summarize the status of relevant Chinese laws and standards for each indicator.

In China, national and local regulations carry immediate force, as they carry legal and market-access consequences. Voluntary standards, while not legally binding, often serve as practical compliance references and are widely adopted in practice. Even draft regulations and policy guidance—at both national and local levels—may shape expectations and signal future directions, prompting companies to align early in order to sustain legitimacy and regulatory goodwill. In this context, the relative scarcity of voluntary safety commitments by Chinese companies may at least in part reflect differences in regulatory expectations and channels for policy engagement.

Below is a high-level summary of how each type of legislation or policy documents influence Chinese AI companies’ behaviors.

National Binding Instruments

What it includes. Foundational statutes passed by the National People’s Congress and its Standing Committee, including the Cybersecurity Law (2017, amended October 2025), Data Security Law (2021), and Personal Information Protection Law (2021) — plus State Council and department rules: Provisions on the Administration of Algorithmic Recommendations in Internet Information Services (2022), Provisions on the Administration of Deep Synthesis of Internet-based Information Services, Generative Artificial Intelligence Interim Measures (2023), Interim Measures for the Management of Generative Artificial Intelligence Services (2023), Regulations on the Management of Network Data Security (2025), Measures for Labelling of Artificial Intelligence-Generated Synthetic Content (2025), Interim Measures for the Administration of Anthropomorphic Artificial Intelligence Interaction Services (2026), among other sectoral rules in finance, healthcare, and autonomous driving.

What it means for companies. National binding instruments are most determinative in driving company behaviors. Statutes generate governance infrastructure and fundamental principles, while departmental rules operationalize concrete implementation measures, including algorithm filings, pre-deployment security assessments, content labeling, training-data provenance logs, and requirements on refusal behaviors.

Enforcement. The Cyberspace Administration of China (CAC), the Ministry of Industry and Information Technology (MIIT), the Ministry of Public Security (MPS), the State Administration for Market Regulation (SAMR), the National Radio and Television Administration (NRTA), and sectoral regulators (e.g. People’s Bank of China, National Financial Regulatory Administration, National Health Commission, National Medical Products Administration, Ministry of Transport).

Service providers face a penalty structure spanning administrative sanctions (e.g. fines up to 50 million RMB or 5 percent of turnover, service suspension, app-store delisting, license revocation), civil liability (e.g. user lawsuits and public-interest litigation), criminal liability (e.g. imprisonment and fines).

Local Binding Instruments

What it includes. Local artificial intelligence regulations enacted by provincial or municipal People’s Congresses and implementing rules issued by local governments, including the Shanghai Regulations on Promoting the Development of the Artificial Intelligence Industry (2022), Shenzhen Special Economic Zone Regulations on Promoting Artificial Intelligence Industry (2022), Beijing Measures on Promoting Innovative Development of Artificial Intelligence (2023), and comparable rules in Hangzhou, Chengdu, and Guangzhou.

What it means for companies. Legally these are predominantly promotional (促进) instruments and are, subordinate to national law — where they conflict, national statutes and departmental rules prevail. They create few new prohibitions; their binding bite is limited to localized duties such as ethics review of high-risk applications, risk classification, and sandbox conditions.

Enforcement. Enforcement is led by local counterparts of the national agencies (local CAC, Economy and Information Technology commissions, Public Security and Market Regulation bureaus) and bodies created by the rules themselves (AI working groups, ethics committees, sandbox administrators). Consequences are mainly incentive-based rather than punitive: companies risk losing subsidies, tax benefits, compute vouchers, and dataset access, or being expelled from sandboxes and pilots.

Voluntary Technical Standard

What it includes. Recommended national standards (GB/T), technical documents, and practice guidelines issued chiefly by the National Information Security Standardization Technical Committee (TC260) under SAMR’s national standardization administration and the guidance of the CAC. Examples include the Basic Security Requirements for Generative AI Services (TC260-003, later GB/T 45654-2025), the Security Specification for Generative AI Pre-training and Fine-tuning Data (GB/T 45652-2025), the Generative AI Data Annotation Security Specification (GB/T 45674-2025), the Assessment Specification for Security of Machine Learning Algorithms (GB/T 42888-2023).

What it means for companies. Legally these standards are recommended (推荐性) and non-binding on their own. Their practical force comes from operationalizing binding regulations: GB/T 45654-2025 supplies the concrete security-assessment methodology providers use to satisfy the filing and assessment duties under the Generative AI Interim Measures, making it a de facto compliance benchmark and reference for regulators and third-party evaluators. [Zou & Zhang, 2025]

Enforcement. As recommended standards they carry no standalone penalties; TC260 and the standardization administration are standard-setters, not enforcers. In practice, conformity is effectively compelled because regulators reference these standards when judging compliance with binding regimes. Therefore, failure to meet a standard surfaces as a failure under the underlying regulation, such as the Interim Measures, rather than as a breach of the standard itself. Consequences therefore flow back through the national binding instruments.

Draft Regulations and Standards

What it includes. Instruments released for public consultation but not yet in force, including draft departmental rules and measures and draft national/industry/municipal standards from TC260 and other bodies (e.g. draft safety, evaluation, and labeling standards before finalization).

What it means for companies. Legally non-binding while in draft — they create no enforceable obligations and cannot be the basis for sanctions. Their significance is anticipatory: they signal the direction and likely content of forthcoming binding rules, give a window to submit comments and shape final text, and let companies pre-position compliance.

Enforcement. Not enforced. There are no penalties for non-conformance with a draft. The relevant bodies (CAC, MIIT, TC260, sometimes the NPC for the AI Law) collect feedback and revise.

Strategic and Policy Guidance Documents

What it includes. High-level, non-binding documents setting national direction, including the State Council’s New Generation Artificial Intelligence Development Plan (2017), the “AI+” initiative guidance, MIIT and NDRC industrial-policy notices, ethical frameworks such as the New Generation AI Ethics Specifications (2021) and the Ethical Norms for New Generation AI, and foreign-facing positions like the Global AI Governance Initiative (2023). These also include high-profile political speeches or party directives from senior leadership.

What it means for companies. No direct legal obligations or liability attached to these documents. Their relevance is strategic: this high-level policy guidance functions as a behavioral steering tool, compelling platform firms to anticipate regulatory trends, publicly align with state priorities, and adjust business practices long before formal laws are enacted.

Enforcement. Not legally enforced. Influence is exercised through the policy and fiscal levers of issuing bodies (State Council, NDRC, MIIT, MOST, CAC, TC260, and the science-and-tech ethics committees): alignment shapes eligibility for subsidies, tax incentives, pilots, and procurement, while divergence risks lost support and reputational/political friction rather than formal sanction.