The Cyberspace Administration of China, in coordination with the Ministry of Industry and Information Technology and the National Development and Reform Commission, on Thursday released the Measures for the Administration of Artificial Intelligence, a 68-article regulatory framework establishing comprehensive governance requirements for AI systems developed, deployed or used within China's jurisdiction.
Scope and Core Obligations
The framework, which enters into force on October 1, 2026, applies to all organisations and individuals developing AI foundation models, deploying AI-generated content services, using AI in consequential decision-making contexts, or operating AI-powered recommendation algorithms. Domestic and foreign technology companies operating AI services accessible from within China will be subject to its full provisions.
Core obligations include mandatory registration and technical disclosure for all AI foundation models with more than one million monthly active users, real-time explainability mechanisms for AI decisions affecting individual rights or significant economic interests, and human oversight requirements for AI systems deployed in healthcare, financial lending, criminal justice, and employment decisions. Companies must maintain comprehensive model cards documenting training data provenance, performance benchmarks and identified failure modes.
The new AI governance framework requires all foundation models with over one million monthly users to register and undergo certification review.
Certification and Safety Testing
A new AI Safety Evaluation Centre, to be established under the auspices of the Chinese Academy of Sciences, will be responsible for conducting mandatory safety evaluations of high-risk AI systems before deployment. The evaluation process will assess systems against a national AI safety standard currently under development, with particular scrutiny applied to systems capable of generating synthetic audio, video, or text content that could be deployed in disinformation campaigns.
Penalties for non-compliance are significant. Companies that deploy uncertified AI systems in restricted categories face fines of up to 10 percent of annual revenue, with senior executives personally liable for fines of up to 500,000 yuan. Repeated or egregious violations can result in operational suspension and cancellation of business licences.
Data Ethics and Algorithmic Fairness
The framework introduces specific provisions addressing algorithmic bias and discriminatory outcomes. AI systems used in recruitment, credit assessment, insurance underwriting and public service allocation must be tested for disparate impact across protected characteristics including gender, ethnicity, age and disability status. Results of bias audits must be disclosed to regulatory authorities and summarised in publicly accessible annual transparency reports.
Data governance provisions require that training datasets used for commercially deployed AI systems be documented with provenance records verifying lawful acquisition, consent where required, and compliance with data localisation requirements. Transfer of AI training data to servers outside China is subject to a security assessment process managed by the Cyberspace Administration.
China's framework establishes human oversight requirements for AI systems deployed in healthcare, finance, criminal justice and employment.
International Context and Diplomacy
The framework's publication is timed to position China as an active governance architect ahead of the G20 Technology Summit in Riyadh, where AI regulation is the centrepiece agenda item. Beijing has expressed support for a multilateral approach to AI governance through the United Nations while simultaneously establishing national frameworks that reflect Chinese values and policy priorities.
The EU's AI Act, which entered application from August 2024, provides the closest comparable reference point. Legal analysts note that while the Chinese and EU frameworks share structural similarities, including risk-tiering, transparency obligations and human oversight requirements, they differ in their treatment of AI used for national security and public order purposes, which are substantially exempted from civilian governance standards under the Chinese framework.
Industry Response
Domestic technology companies including Baidu, Alibaba Cloud and Huawei's AI division expressed support for the framework, noting that regulatory clarity removes competitive uncertainty and provides a stable basis for long-term investment in AI infrastructure. Overseas companies operating in China, including Microsoft Azure and Google Cloud's locally licenced services, confirmed they were reviewing compliance obligations and expected to be operationally compliant by the October implementation date.
Academic observers have noted that China's rapidly maturing AI governance architecture, building on earlier interim measures for generative AI and algorithmic recommendation systems, now represents one of the world's most comprehensive regulatory frameworks for artificial intelligence and could serve as a model for jurisdictions seeking to balance innovation with accountability.