China AI Ethics Review Measures

Overview

The Measures for the Ethical Review of and Services for Artificial Intelligence Science and Technology (Trial) establish an AI-specific system for ethical governance of scientific research, technology development and other covered activities. Notice 工信部联科〔2026〕75号 was jointly issued by the Ministry of Industry and Information Technology, National Development and Reform Commission, Ministry of Education, Ministry of Science and Technology, Ministry of Agriculture and Rural Affairs, National Health Commission, People's Bank of China, Cyberspace Administration of China, Chinese Academy of Sciences and China Association for Science and Technology. Article 1 connects responsible innovation and healthy industrial development with fairness, justice, harmony and safety. Review obligations are combined with services for identifying and controlling ethical risks. The State Council policy publication contains the issuing notice and official attachment.

Article 2 covers activities conducted within the People's Republic of China that may create ethical risks affecting human dignity, public order, life and health, the ecological environment or sustainable development. It also covers other scientific activities requiring AI ethics review under applicable legislation and national provisions. This is a risk-dependent review framework rather than blanket licensing for every AI product or use. Article 37 makes the measures effective from issuance on March 20, 2026. Publication on the MIIT website followed on April 2, 2026.

Definitions

The operative scope is defined through covered AI scientific and technological activities rather than a general technical definition of artificial intelligence. Article 3 requires ethical considerations throughout an activity's lifecycle. Its principles include human welfare, respect for life and rights, fairness and justice, reasonable risk control, openness and transparency, privacy and security, and controllability and trustworthiness. These principles guide procedural classification and substantive review; they are qualitative review considerations rather than numerical risk-scoring thresholds.

An AI science and technology ethics committee is an institution's review body. An AI ethics review and services center is a specialized body established with local or departmental support. Article 33 defines “local” as the provincial management department designated by the provincial people's government to oversee AI ethics review and management; “relevant competent departments” means relevant State Council departments. Unqualified deadlines are calendar days. The expert re-review list identifies activities requiring an additional review stage, may be dynamically adjusted, and operates alongside the broader scope of initial ethics review.

Governance and Institutional Framework

Article 9 makes universities, research institutions, medical and health institutions, enterprises and other participating organizations responsible for managing their AI ethics review. They must establish committees in accordance with Article 4 of the general science and technology ethics review measures, provide personnel, premises and funding, and protect independent operation. Article 10 incorporates the general instrument's committee governance requirements and requires expertise in AI technology, applications, ethics and law. Ethics management system certification is encouraged for qualified organizations, rather than required universally.

Article 11 permits local and relevant competent departments to establish service centers for commissioned review, expert re-review, training and consultation. A center must maintain procedures, qualified full-time staff and supervisory accountability, and cannot conduct both initial review and expert re-review for the same activity. Under Article 29, the Ministry of Science and Technology coordinates nationwide ethics regulation, while MIIT and relevant departments undertake AI ethics governance. Departments oversee their industries and institutional systems; designated provincial bodies oversee their territories. Article 36 assigns interpretation of the measures to MIIT and relevant departments.

Key Focus Areas

Article 15 establishes six substantive review dimensions. Human welfare addresses scientific and social value, sustainable development and a reasonable balance between risks and benefits. Fairness and justice address training-data selection and algorithm, model and system design, including prevention of bias, discrimination and algorithmic exploitation, and objective, inclusive allocation, access and decision-making. Controllability and trustworthiness address robustness in open environments, extreme conditions and disruptive circumstances; users' ability to control, guide and intervene in basic operations; continuing monitoring; and arrangements for unexpected situations.

Transparency and explainability concern appropriate disclosure of purposes, operating logic, interaction instructions and potential risks, together with effective technical measures supporting explainability. Accountability and traceability concern logs and other records sufficient to track and manage data, algorithms, models and systems throughout the chain, as well as personnel qualifications. Privacy protection covers data collection, storage, processing and use, including research into new data technologies. Articles 4–8 support standards development, monitoring and assessment services, certification, consultation, research tools, public education and professional training, with additional support for small, medium and micro enterprises.

Implementation Framework

Under Article 12, the activity leader applies to the institution's committee. Where no committee exists or it cannot competently review the activity, the institution commissions a service center. Unaffiliated individuals must commission a qualified service center. Applications principally include the activity plan, institutional qualifications, personnel, funding, algorithm mechanisms, data sources and acquisition methods, testing methods, intended products, applications and populations. An ethical risk assessment, prevention and emergency plans, and an ethics and research-integrity commitment are also required. Article 13 requires an acceptance decision, notification, procedure selection and a complete, one-time notice of missing materials.

General review meetings require at least five committee members covering the prescribed expertise. Advisory experts without direct interests may advise but cannot vote. Article 16 requires a decision within 30 calendar days after acceptance, with an expressly defined extension for complex cases or necessary material corrections. Simplified review uses at least two designated members for qualifying low-risk activities, minor approved-plan changes that do not increase the risk-benefit ratio, or follow-up reviews without major prior adjustments. Negative opinions, doubts or disagreement require escalation to general review. Article 18 permits participating organizations in multi-institution cooperation to recognize each other's review results according to actual circumstances.

Three categories require expert re-review after preliminary approval: human-machine integration with strong human effects; systems capable of public-opinion mobilization or social-awareness guidance; and highly autonomous decision-making in settings involving safety or health risks. Applications are routed through the responsible department or provincial body, with the lead institution applying for joint activities. Expert re-review assesses the compliance and reasonableness of preliminary opinions, with feedback within 30 calendar days of the application. Emergency review generally finishes within 72 hours, while preliminary review before required expert re-review generally finishes within 36 hours.

Monitoring and Evaluation

Article 17 requires activity leaders to identify changes in ethical risk promptly and report them to the reviewing committee or service center. Approved activities undergo follow-up review, generally at intervals not exceeding 12 months. Reviewers may suspend or terminate activities when necessary. Article 25 strengthens oversight of activities on the expert re-review list, generally limiting intervals to six months. Major ethical risk changes trigger renewed ethics review and an application for expert re-review. Approval therefore creates a continuing governance relationship responsive to changing activity conditions.

Article 30 requires registration of committee information and listed activities through the National Science and Technology Ethics Management Information Registration Platform under Articles 43–45 of the general ethics measures. Organizations submit previous-year committee work reports and implementation reports for listed activities. Service centers also register and submit previous-year work reports. The Ministry of Science and Technology and relevant competent departments share registration information synchronously. Article 31 requires accessible channels for reporting suspected ethics violations. Article 28 emphasizes follow-up and process supervision for emergency reviews. Oversight centers on review, registration, reporting and supervision rather than public scoring or ranking.

Penalties, Liability, and Appeals

Article 32 directs investigation, handling and penalties for violations to the Cybersecurity Law, Data Security Law, Personal Information Protection Law, Science and Technology Progress Law and other applicable provisions. Legal consequences depend on the conduct and competent authority, including violations during AI scientific activities and ethics-related work. The measures rely on those legal frameworks rather than creating a standalone fine schedule or criminal offence. Suspension or termination during follow-up review is a protective governance measure distinct from a statutory fine.

Article 16 establishes an internal challenge procedure. Decisions requesting modification must include recommendations, and refusals must include reasons. An applicant dissatisfied with a decision must appeal to the committee or service center within three working days after delivery. Where the grounds are sufficient, the body must issue a new decision within seven working days. These expressly working-day deadlines differ from calendar-day review periods. Other applicable law governs judicial review and compensation, while the measures establish the internal reconsideration route.

Relationship to Other Instruments

Article 1 bases the measures on the Science and Technology Progress Law, the Opinions on Strengthening Science and Technology Ethics Governance, and the Measures for Ethical Review of Science and Technology (Trial). The general measures remain foundational. The AI instrument incorporates their rules on committee organization, follow-up review, expert re-review materials and procedures, and registration and reporting. Article 35 directs matters outside the AI measures to the general measures and relevant legislation, and preserves compatible special departmental provisions for particular industries or institutional systems.

Article 26 addresses overlap with deep synthesis, algorithm recommendation and generative AI service regulation. Separate expert re-review may be omitted where applicable registration, filing, administrative approval or comparable supervision incorporates ethics compliance as an approval condition or regulatory matter. This conditional relief applies to the additional expert stage, while initial review, monitoring and substantive ethics obligations continue. Article 34 permits local and departmental implementing measures and detailed rules, and allows scientific and technological social organizations to develop sector-specific review and service specifications and guidance.

International Alignment

Article 4 supports international, national, industry and association standards and platforms for international standardization exchange. Universities, research institutions, medical institutions, enterprises and scientific and technological social organizations are encouraged to develop, validate and promote ethics standards. Article 5 calls for international AI ethics exchange and cooperation. These provisions promote capacity building and professional dialogue while maintaining domestic review requirements.

The territorial scope remains activities conducted within China. Article 18 permits review-result recognition in multi-institution cooperation according to actual circumstances, while Article 35 retains the general ethics measures for matters outside the AI-specific rules, including relevant international research arrangements. International standardization cooperation is distinct from approval equivalence: foreign ethics processes, overseas certifications and internationally developed assessment tools do not automatically substitute for domestic review. The instrument's governance framework concerns covered domestic activities rather than a separate overseas enforcement regime.

Implementation Timeline

Sources and References