China sets global first with pre-development AI ethics mandate

China has instituted a groundbreaking AI ethics review framework that mandates ethics assessments before any AI development begins, including prior to model pre-training, a stage earlier than any Western jurisdiction currently requires. Formalized in the Administrative Measures for the Ethical Review and Services of Artificial Intelligence Science and Technology (Trial) effective April 3, 2026, this pre-development review shifts governance from the traditional pre-market focus to an anticipatory, upstream intervention, signaling a proactive approach to AI risk management.

The framework employs a distributed, multi-tiered oversight model that integrates internal ethics committees within every AI-engaged organization—ranging from universities and research institutions to medical organizations and companies—with accredited external ethics review service centers to support entities lacking internal capacity. For projects deemed high-risk, a mandatory second review by government-assigned expert panels ensures an additional layer of scrutiny, reflecting a comprehensive governance architecture designed to balance institutional responsibility with centralized oversight.

China’s ethics review process is not only mandatory but also rigorously time-bound and continuous, requiring decisions within 30 days of submission or a formal extension notice. Approved AI projects are subject to ongoing monitoring with follow-up reviews scheduled at intervals no longer than 12 months for ordinary cases and six months for high-risk projects, underscoring a commitment to sustained oversight rather than one-off approvals, thereby institutionalizing compliance and accountability throughout the AI lifecycle.

China has established an extensive and legally backed framework of nearly 200 formalized AI standards that span the entire AI system lifecycle, including model safety classification, data governance, and sector-specific deployment guidance across industries such as healthcare, finance, transportation, and manufacturing. This comprehensive corpus, developed under the Ministry of Industry and Information Technology, aims to enhance interoperability and set clear governance expectations, reflecting a deliberate effort to integrate ethical oversight with technical standardization to balance innovation with accountability.

To operationalize these standards, China has launched city-level pilot programs that uniquely mandate pre-development AI ethics reviews, requiring all universities, research institutions, medical organizations, and companies engaged in AI to establish internal ethics committees and submit projects for review before any development or model pre-training begins. This early-stage governance model, unparalleled globally, shifts the focus from post-development or pre-market assessments to proactive ethical scrutiny at the project planning phase, enabling authorities to better assess social and operational implications across diverse sectors.

China’s AI ethics review system employs a distributed, three-tier governance structure that combines internal ethics committees, accredited external review centers, and government expert panels for high-risk projects, ensuring layered oversight and compliance with the nearly 200 formalized standards. This multi-level approach not only delegates responsibility within organizations but also integrates external and governmental scrutiny, creating a robust framework designed to manage risks effectively while fostering responsible AI innovation at scale.

China’s AI regulatory approach stands apart globally by mandating ethics reviews at the pre-development stage, requiring evaluation before model pre-training begins. This early-stage governance contrasts sharply with the US’s market-driven, fragmented system and the EU’s risk-based, post-development conformity assessments, which typically evaluate finished AI systems. By instituting a mandatory pre-R&D review, China intervenes fundamentally earlier in the AI lifecycle, enabling planned compliance and proactive risk mitigation.

China’s multi-tiered, distributed oversight model—comprising internal ethics committees, accredited external review centers, and government-assigned expert panels for high-risk projects—creates a layered governance architecture distinct from the more centralized or downstream regulatory approaches seen in the US and EU. This structure supports continuous monitoring and follow-up reviews, reinforcing ongoing compliance beyond a one-time pre-market check, and reflects a shift from aspirational policy to operational enforcement.

The mandatory 'release gate' mechanism exemplifies China’s unique emphasis on pre-release controls, requiring government security assessments and algorithm filings before any public deployment. Unlike the US, which relies on voluntary frameworks and export controls that preserve legal deniability, China’s transparent, prescriptive system defines clear release evidence that companies must assemble, as demonstrated by the 346 generative AI services filed with the Cyberspace Administration of China by April 2025. This approach enables companies like Baidu to launch products such as Ernie Bot swiftly and compliantly.