China's medical device GMP takes effect as AI diagnostic rules stay in draft

Vendors selling diagnostic AI into China now face two separate gates, product registration and data compliance, and the EU adds a further layer through MDR classification and AI Act duties. Buyers should expect longer clinical validation and more documentation before deployment. The two NMPA guidelines are still drafts, so the final requirements are not yet fixed.

China's two gates

China treats AI software that diagnoses, triages or guides treatment as a medical device. It must be registered with the National Medical Products Administration before sale or hospital use.

The second gate is data. Medical AI must follow the Personal Information Protection Law, the Data Security Law and the Cybersecurity Law, which govern the patient data used to train or run a model. Miss either gate and a launch stalls. In mid-September 2026 the NMPA published two drafts: a Guideline on Artificial Intelligence Medical Devices and a Guideline on Clinical Evaluation of Artificial Intelligence Multi-Disease Auxiliary Decision-Making Medical Device. Analysts read the pair as a layered system, with a horizontal lifecycle and algorithm framework plus vertical clinical evaluation rules for complex products. Both remain drafts.

Classification decides the route. AI-assisted diagnosis software sits under code 21-04-02, and foreign manufacturers cannot register directly, so they must appoint a legal agent in China.

Why the rules matter now

China's approval record shows where diagnostic AI lands. A 2026 analysis counted 154 AI device approvals from 2020 through June 2025, rising from 9 in 2020 to 45 in 2024. Class III, the highest risk class, accounted for 79.9% of them.

Demand is pushing the queue. openPR, citing a market report, estimates the global AI in medical diagnostics market will grow at around 22% over the next five years. It points to rising chronic and lifestyle-related disease and shortages of skilled medical professionals as drivers of adoption. For vendors, the review route rather than the model sets the launch date.

Europe's parallel path

The EU splits the question differently. Researchers from HUN-REN SZTAKI, Semmelweis University and Pázmány Péter Catholic University developed a proof-of-concept machine-learning system for premature infants. It assesses positioning from continuous video in neonatal intensive care units, and the study appeared in Pediatric Research in June 2026.

Under Article 2(1) of the EU Medical Devices Regulation, a clinical version would likely qualify as medical device software, according to Taylor Wessing. Under Rule 11 of Annex VIII, software providing information used to take diagnostic or therapeutic decisions is generally Class IIa. It rises to Class IIb or III where the consequences may be more serious. If it also counts as an AI system under the AI Act, high-risk requirements would generally apply from 2 August 2028.

Article 61 of the MDR requires enough clinical evidence to show conformity with safety and performance requirements. The researchers asked for multicentre validation and prospective studies before clinical adoption. Their study covered stable premature infants at two NICUs within one institution and excluded more complex cases.

What happens next

China's revised GMP is already in force, so quality-system documents for AI-enabled devices now face the new requirements. The two NMPA guidelines from mid-September 2026 remain drafts. In Europe, the AI Act's high-risk obligations for covered systems apply from 2 August 2028. For now, the paperwork is the part with fixed dates.