China: NMPA YY/T 2029-2026 EEG Dataset Quality Standard for AI BCI

On September 14, 2026, China's National Medical Products Administration (NMPA) published Announcement No. 79 of 2026, releasing the medical-device industry standard YY/T 2029-2026, "Brain-Computer Interface Medical Devices: Quality Requirements and Evaluation Methods for EEG Datasets Used in AI Algorithms." The standard sets the first dedicated quality benchmark for the EEG data that trains AI models in brain-computer interface (BCI) medical devices, and it applies from September 1, 2027.

NMPA issued the standard under the State Council medical-device framework (Order No. 739, 医疗器械监督管理条例), and it follows the agency's recent BCI classification guidance. For the manufacturers it targets, the practical effect is that the EEG datasets feeding their AI algorithms, the methods used to evaluate those datasets, and the surrounding collection, annotation and storage controls now have a published reference to meet at registration.

What does YY/T 2029-2026 require, and is it mandatory?

The "YY/T" prefix marks a recommended medical-device industry standard: the "/T" denotes 推荐 (recommended), as opposed to a mandatory "YY" or national "GB" standard. YY/T 2029-2026 fixes quality requirements and evaluation methods for the EEG datasets used to train or validate AI algorithms in BCI medical devices, covering dataset collection, annotation, storage and quality control, plus the methods a manufacturer uses to show a dataset is fit for its algorithmic purpose.

Recommended does not mean optional in practice. Under Order No. 739, NMPA's Center for Medical Device Evaluation (CMDE) assesses conformity against recognised standards during registration review, and a published YY/T on dataset quality is the reference CMDE reviewers will reach for when examining BCI-AI products. Teams should treat alignment as expected evidence, not a courtesy. The standard number, name, scope and implementation date are set out in the announcement's annex (medical-device industry standard information table).

Who must align, and which products are in scope?

The standard applies to BCI medical devices that use AI algorithms operating on EEG data, whether invasive or non-invasive. The exposed audience is concrete: Chinese BCI developers building or registering AI-EEG products, including Neuracle (博睿康), which holds the Class III NEO-ONESCI invasive system, BrainCo, NeuroXess and StairMed, together with foreign neurotech companies seeking NMPA registration for the China market.

For these manufacturers, the dataset is now a regulated input. Training data, clinical-evaluation data and any EEG corpora used to validate the algorithm must be collected, annotated and stored in line with YY/T 2029-2026, and the standard's evaluation methods define how dataset fitness is demonstrated to a reviewer. Products already in CMDE review, or preparing a Product Technical Requirements (PTR) submission, should map their existing data governance against the new requirements now.

How does the standard fit China's medical-device regime?

China regulates medical devices under State Council Order No. 739 on a three-class risk pathway: Class I filing, Class II provincial registration, and Class III (plus all imports) central NMPA registration with CMDE evaluation. BCI devices that run AI on neural signals generally fall in Class III, bringing central registration, Product Technical Requirements, clinical evaluation, Unique Device Identification and post-market surveillance.

YY/T 2029-2026 slots into this regime as the dataset-layer companion to the classification and PTR framework: Order 739 sets the class a BCI-AI device sits in and how it is registered, while the new standard tells reviewers what good EEG data for its AI algorithm looks like. SAMR's draft Medical Device Administration Law (11 chapters, 190 articles, a 2026 legislative priority) would later elevate the regime from administrative regulation to a full NPC statute, but the YY/T standard applies on its own September 1, 2027 date regardless of that track. The full announcement is on the NMPA medical-device standards announcements page.

What should teams do before September 1, 2027?

First, confirm whether your product is in scope: any BCI medical device that runs an AI algorithm on EEG data for the China market is the target class. Second, audit your dataset lifecycle against the standard's collection, annotation, storage and QC requirements, and document the gaps. Third, update your PTR and CMDE submission evidence to reference YY/T 2029-2026 conformity, including the evaluation methods used to demonstrate dataset fitness. Finally, brief regulatory affairs, the algorithm team and quality together, because the standard touches all three functions and the 2027 deadline leaves little room for a sequential handoff.

Continuous, per-jurisdiction real-time monitoring surfaces a development like this the moment NMPA publishes it, before the standard reaches a Western tracker. Obsidian tracks Chinese medical-device standards as they post, so the teams above can begin alignment on day one.

Subscribe to the free newsletter

To close: verify whether your BCI-AI product is in scope, mark the September 1, 2027 implementation date against your registration timeline, and brief regulatory affairs, algorithm and quality leads this week. Obsidian will keep watch as NMPA moves from publishing the standard to applying it in CMDE review.