Establishes ethics review committees for S&T activities, including AI, to assess ethical risks. Requires committees to follow principles of independence and transparency. Dictates emergency review procedures and mandates expert reconsideration for high-risk AI activities. Specifies penalties for violations.
Analysis summaries, actor details, and coverage mappings were LLM-classified and may contain errors.
This is a binding regulatory instrument with mandatory language throughout ('shall', 'must'), explicit enforcement mechanisms including penalties and sanctions, and formal administrative oversight structures. The document establishes legally enforceable obligations for S&T ethics reviews with clear consequences for non-compliance.
The document has good coverage of approximately 8-10 subdomains, with strong focus on governance failure (6.5), AI system security (2.2), privacy compromise (2.1), lack of transparency (7.4), goal misalignment (7.1), dangerous capabilities (7.2), and lack of robustness (7.3). Coverage is concentrated in governance structures, privacy/security, and AI safety domains, with minimal coverage of discrimination, misinformation, or socioeconomic impacts.
The document governs S&T activities across multiple sectors, with explicit coverage of Scientific Research and Development Services, Health Care and Social Assistance, Educational Services, and Information sectors. It applies to institutes of higher education, scientific research institutions, medical and health institutions, and enterprises engaged in life sciences, medicine, and artificial intelligence.
The document comprehensively covers all stages of the AI lifecycle, with particularly strong emphasis on Plan and Design (through ethical risk assessment requirements), Verify and Validate (through the detailed review procedures), and Operate and Monitor (through follow-up review requirements). It addresses the entire lifecycle from initial planning through ongoing operational monitoring.
The document explicitly mentions AI systems and activities, with specific focus on high-risk AI applications including algorithmic models, automated decision-making systems, and human-computer fusion systems. It does not use terms like 'frontier AI', 'general purpose AI', 'foundation models', or 'generative AI', but addresses AI broadly through references to 'artificial intelligence' and specific AI applications. No compute thresholds or open-weight model distinctions are mentioned.
Ministry of Science and Technology; Chinese central government
The document is formulated by the Ministry of Science and Technology under the authority of the Chinese central government, as indicated by references to national laws and the Ministry's responsibility for interpretation and platform construction.
Ministry of Science and Technology; local oversight departments; provincial-level S&T administrative departments; relevant main industrial oversight departments; agencies with jurisdiction
The document designates multiple levels of government agencies with enforcement authority, including the Ministry of Science and Technology, local and provincial-level departments, and industrial oversight departments with powers to impose penalties, conduct investigations, and pursue legal liability.
S&T ethics (review) committees; Ministry of Science and Technology; local and relevant main industrial oversight departments; work units
The document establishes S&T ethics review committees at work units as primary monitoring bodies, with oversight from the Ministry of Science and Technology through a national registration platform, and supervision by local and industrial oversight departments.
Institutes of higher education; scientific research institutions; medical and health institutions; enterprises
The document applies to organizations conducting S&T activities including AI development and deployment, specifically naming institutes of higher education, scientific research institutions, medical and health institutions, and enterprises engaged in life sciences, medicine, and artificial intelligence.
12 subdomains (8 Good, 4 Minimal)