AI Ethics Review and Committees
China’s ethics review system: mandatory internal committees, optional regional support, and expert re-review escalation for disputed or high-risk cases.
Science and Technology Ethics Review Measures (Trial)
The Measures for Ethical Review and Service of Artificial Intelligence Science and Technology (Trial) (人工智能科技伦理审查与服务办法(试行)), issued jointly by 10 government departments including MIIT and MOST, took effect in April 2026. MOST leads national ethics supervision while MIIT coordinates AI-specific ethics governance and interprets these measures. This regulation applies specifically to AI scientific and technological activities that may pose ethical risks, building on the broader 2023 Science and Technology Ethics Review framework.
The regulation requires universities, research institutes, medical institutions, and enterprises engaged in AI research or deployment to establish internal ethics committees. These committees must evaluate all AI activities against six defined ethical dimensions before development or deployment may proceed.
Unlike the GenAI Interim Measures (生成式人工智能服务管理暂行办法, effective August 2023) or the CAC filing system, the Ethics Review Measures apply to AI research and technology activities broadly, not only to public-facing services. The obligations fall on the organization conducting the AI activity, not the end user.
Three-Tier Ethics Review System
Click each tier to see the review requirements and escalation triggers.
Every covered organization must establish its own ethics committee. The committee conducts the initial ethics review of all AI activities before they proceed.
- Committee must include experts in AI technology, ethics, and law
- Minimum five members required per review meeting
- Committee independence and dedicated funding are mandated
- Applies to universities, research institutes, medical institutions, and enterprises
Local governments may establish ethics review service centers to support organizations that lack sufficient internal review capacity or to handle cases that exceed the scope of internal committees.
- Established by local government authorities
- Provides review services for organizations in the jurisdiction
- A service center that reviews a project at this tier cannot also serve as the re-review body for the same project
- Must respond within 30 days of receiving a review request
When an organization or service center review produces a disputed result, or when the AI activity falls into a high-risk category, an independent expert panel conducts a re-review.
- External experts advise but do not vote on outcomes
- Panel must be independent from the original reviewing body
- Six-month follow-up period after the re-review decision
- High-risk AI categories (Annex) trigger mandatory escalation
How Does This Compare to Western Ethics Boards?
China’s system mandates internal committees and provides government-backed escalation paths for complex cases. The TJS 8-stage committee framework offers a parallel governance structure used by multinational organizations.
Six Mandatory Review Dimensions
Every AI ethics review must assess the activity against all six dimensions. Click each dimension to see evaluation criteria.
Human Well-Being
Value alignment, risk-benefit analysis, and long-term sustainability of the AI activity.
- Does the AI activity serve identifiable human or societal value?
- Are the benefits proportionate to the risks imposed on individuals?
- Is the system designed with long-term environmental and social sustainability in mind?
- Does the activity avoid causing irreversible harm to physical or mental well-being?
Fairness and Justice
Bias prevention, anti-discrimination, and equitable access to AI-driven outcomes.
- Has the training data been assessed for demographic or socioeconomic bias?
- Does the algorithm prevent exploitation of vulnerable populations?
- Are system outputs equitable across different user groups?
- Is there a documented process for detecting and correcting algorithmic discrimination?
Controllability and Trustworthiness
System reliability, user control mechanisms, and contingency plans for failure conditions.
- Can the system be stopped, overridden, or rolled back by a human operator?
- Are there defined contingency plans for unexpected system behavior?
- Has the system been tested for adversarial inputs and edge cases?
- Are performance degradation thresholds documented with escalation procedures?
Transparency and Explainability
Clear disclosure of the system’s purpose, decision logic, and known risk factors.
- Is the purpose of the AI system disclosed to affected individuals?
- Can the decision logic be explained in terms understandable to non-experts?
- Are known risks and limitations documented and communicated?
- Is there a mechanism for affected parties to request a human review of automated decisions?
Responsibility and Traceability
Full-chain audit logging, named accountability, and incident response procedures.
- Is every decision in the AI pipeline logged with sufficient detail for post-incident review?
- Are named individuals or roles accountable for each stage of the AI lifecycle?
- Can the provenance of training data be traced back to its source?
- Is there an incident response plan with defined notification timelines?
Privacy Protection
Data minimization, consent management, and safeguards for personal information in AI systems.
- Does the system collect only the data strictly necessary for its stated purpose (PIPL Art. 6)?
- Is user consent obtained before processing sensitive personal information?
- Are data anonymization or de-identification techniques applied where feasible?
- Does the system comply with cross-border data transfer requirements under PIPL Art. 38?
Map These Dimensions to International Standards
The six review dimensions parallel requirements in ISO 42001, NIST AI RMF, and the EU AI Act. TJS maps Chinese ethics review obligations against global frameworks.
AI-Relevant High-Risk Categories from the Annex
The Annex covers high-risk science and technology activities broadly (including life sciences, medicine, and other fields). The three categories below are the AI-relevant portions extracted from the full list. Activities in these categories require escalated review. Click each card to see examples.
Human-Machine Fusion
Systems that affect human behavior, emotions, or health through direct interaction or physiological integration.
This category covers AI systems with a direct physiological or psychological nexus to the human user.
- Brain-computer interfaces and neural signal processing
- AI-driven mental health intervention systems
- Emotion recognition used in behavioral modification
- AI-assisted medical devices that adjust treatment in real time
Public Opinion Mobilization
Systems with the capacity to shape, direct, or amplify public opinion at scale.
This category targets AI systems that can generate, organize, or recommend content at a scale that influences public discourse.
- Algorithm-driven news recommendation engines
- AI content generation at social media scale
- Automated public comment or review systems
- Deep synthesis (deepfake) creation platforms with broad distribution
Overlaps directly with the Algorithm Recommendation Provisions (2022) and Deep Synthesis Provisions (2023) filing requirements.
Autonomous Decision Systems
Highly autonomous systems making decisions in scenarios involving public safety or individual health.
AI systems that make consequential decisions with limited or no human oversight in safety-critical domains.
- Autonomous vehicles and traffic management systems
- AI-driven medical diagnosis without physician confirmation
- Critical infrastructure control systems (power grids, water treatment)
- Automated safety monitoring and emergency response systems
Ethics Committee Composition Requirements
The regulation specifies both the expertise areas and the operational independence requirements for ethics committees.
AI Technology Experts
Members with direct expertise in the AI technologies under review. Must be able to evaluate technical feasibility, system architecture, and performance characteristics.
RequiredEthics Scholars
Members with academic or professional background in applied ethics, bioethics, or technology ethics. Responsible for evaluating the six mandatory review dimensions.
RequiredLegal Experts
Members with expertise in data protection law, cybersecurity regulation, and AI-specific compliance obligations. Must assess alignment with PIPL, DSL, CSL, and sector-specific rules.
RequiredExternal Advisors
Subject matter experts from outside the organization who provide guidance on specialized topics. External advisors may participate in review discussions but do not vote on outcomes.
Advisory OnlyA global ethics board is unlikely to satisfy the Chinese requirement on its own. The regulation mandates institutional-level oversight within the organization conducting the AI activity in China.
- Foreign-invested enterprises (WFOEs, JVs) should establish a China-specific committee or sub-committee that reports locally.
- The regulation does not explicitly require Chinese nationality for members, but committee proceedings and review documentation should be in simplified Chinese.
- External ethics scholars are expected to come from recognized academic or research institutions. Verify whether foreign-institution affiliations are accepted by your provincial authority.
Building a Committee from Scratch?
The TJS Committee Hub covers the full 8-stage process for standing up an AI governance committee, from charter drafting through operational maturity.
Review Deadlines and Follow-Up Periods
The regulation specifies mandatory response timelines at each tier of the review system.
The 30-day response requirement applies to service center and expert re-review tiers. The 6-month follow-up period runs from the date of the final review decision, during which the organization must demonstrate that review conditions are being met.
Who Must Establish an Ethics Committee?
The regulation specifies four categories of organizations. Scroll horizontally on mobile.
| Organization Type | Internal Committee | Ethics Review Required | High-Risk Escalation | Follow-Up Reporting |
|---|---|---|---|---|
| Universities (高校) | ✓ | ✓ | ✓ | ✓ |
| Research Institutes (研究所) | ✓ | ✓ | ✓ | ✓ |
| Medical Institutions (医疗机构) | ✓ | ✓ | ✓ | ✓ |
| Enterprises (企业) | ✓ | ✓ | ✓ | ✓ |
| Local Govt. Service Centers | – | ✓ | * | ✓ |
* A service center that performs a Tier 2 review cannot also serve as the re-review body for the same project at Tier 3.
The obligation to establish an internal ethics committee falls on the organization conducting the AI research or deployment activity. For multinational companies operating AI systems within China, this means that the China-based legal entity must maintain its own committee, even if the parent company has a global AI ethics board.
Mapping China’s Ethics Review to the TJS Framework
How the China-specific requirements map to the TJS 8-stage committee implementation model.
| China Ethics Review Requirement | TJS Framework Stage | Key Alignment |
|---|---|---|
| Internal committee establishment | Stage 1: Charter | Committee formation, governance structure, funding mandate |
| Multi-domain expertise (tech, ethics, law) | Stage 2: Composition | RACI matrix, named roles, expertise requirements |
| Six-dimension evaluation framework | Stage 4: Risk Assessment | Risk tier classification, multi-framework mapping |
| Three high-risk categories (Annex) | Stage 5: Risk Treatment | Harm taxonomy, adversarial testing, risk mitigation |
| Escalation to service center / expert panel | Stage 6: Monitoring | Incident escalation procedures, human oversight |
| 30-day response, 6-month follow-up | Stage 8: Review Cadence | Board reporting, periodic reassessment |
Get the Full 8-Stage Implementation Guide
The TJS implementation guide covers each stage with ISO 42001, NIST AI RMF, and EU AI Act mapping, plus downloadable tools at every stage.
Need Help Setting Up an Ethics Review Committee?
TJS advisors help organizations establish ethics review committees, define review dimensions, and build escalation workflows that meet the April 2026 requirements.
Talk to a TJS Advisor →