China Issues World's First AI-BCI EEG Data Standard as FDA and EMA Lag
核心洞察
China's NMPA approved the world's first product standard for AI-powered brain-computer interface (搜索) medical devices processing EEG data, effective September 1, 2027.
The standard sets uniform technical requirements and testing methods across five EEG data stages: collection, processing, labeling, storage, and access.
No FDA, EMA, or ISO equivalent addresses EEG training-data quality, though FDA issued draft AI device guidance in January 2025 and ISO/IEC TS 27571:2026 covers data format.
China's National Medical Products Administration (搜索) (NMPA) approved and issued the world's first product standard for brain-computer interface (搜索) (BCI) medical devices that use artificial intelligence to process electroencephalogram (搜索) (EEG) data, the agency announced Monday. The standard takes effect September 1, 2027, and establishes uniform technical requirements and testing methods across the full lifecycle of BCI medical device data: acquisition, processing, annotation, storage, and access.
The NMPA said BCI medical devices function as intelligent machines that can "read" brain signals. By learning from large volumes of EEG data and training AI algorithms, they can accurately interpret patients' intentions. The quality and reliability of that data are critical not only to treatment effectiveness but also to patient safety.
To support major innovations in China's BCI medical device sector and meet urgent industry development needs, the NMPA adopted an expedited process for developing and approving the standard. According to the agency, it provides a unified quality benchmark for EEG datasets used in the research, development, and manufacturing of BCI medical devices, reaching a world-leading level.
A Data Quality Bottleneck for AI-Driven Neurodevices
The standard targets a specific technical problem. For an AI model to reliably decode a patient's intended action from brain signals, it must be trained on large volumes of labeled EEG data — recordings that tell the model what a signal looked like when a patient imagined a specific movement. The quality of that training data is the primary bottleneck in AI-BCI performance.
If data is collected under inconsistent conditions — differing electrode placements, sampling rates, or impedance levels — labeled carelessly, or stored without adequate metadata, the resulting model may perform well in controlled testing but fail in real clinical use. In that scenario, a misread instruction from a spinal cord injury (搜索) patient's brain could translate into a mismovement causing a fall or injury.
The NMPA standard is designed to address common sector challenges, including inconsistent dataset quality control criteria and non-standardized dataset construction processes. By mandating uniform quality requirements across every stage of the EEG data lifecycle, it establishes the technical floor that AI-BCI developers must clear before their training datasets can be considered valid.
Regulatory Asymmetry With the FDA and International Bodies
No other regulatory body has issued a standard specifically addressing EEG training-data quality for BCI devices. The FDA's draft guidance on AI-enabled medical device software functions, published in January 2025, covers lifecycle management, training data practices, and performance evaluation across AI medical devices generally, but does not establish specific quality requirements for EEG training datasets used in BCI devices. The FDA's BCI device guidance finalized in May 2021 covers non-clinical testing and clinical considerations for paralysis and amputation applications.
At the international level, the ISO BCI data format standard ISO/IEC TS 27571:2026 specifies a standardized data format for non-invasive BCI technologies including EEG, defining how data should be structured, what metadata should accompany it, and what naming conventions should be used. A data format standard is not a quality standard: it tells developers how to arrange data, not what the data must contain, how it must be validated, or what accuracy requirements a labeled dataset must meet for AI model training. China's standard addresses those latter questions.
In November 2025, UNESCO adopted a neurotechnology ethics recommendation addressing mental privacy, informed consent, and protection from coercion. That framework is non-binding, establishing principles for governments and companies to consider, whereas China's EEG standard is a mandatory, technically specific, legally enforceable product requirement.
The practical consequence is that Chinese BCI companies developing AI-powered devices can plan their data pipelines against a known, specific technical standard, while companies in the United States and Europe developing comparable devices must anticipate requirements their regulators have not yet specified.
Building on a Broader BCI Framework
The EEG standard is the third BCI-specific industry standard China has issued and the latest component of a regulatory architecture assembled over roughly three years. In June 2025, the Ministry of Industry and Information Technology (搜索) established a dedicated BCI standardization committee. Two months later, seven-ministry BCI development guidelines were jointly issued with binding targets: key technological breakthroughs by 2027 and two to three globally competitive Chinese BCI enterprises by 2030.
In March 2026, the NMPA granted commercial approval to the NEO system (搜索), a coin-sized epidural implant developed by Shanghai-based Neuracle Technology (搜索) in partnership with Tsinghua University, making it the world's first invasive BCI approved for commercial medical use. Within roughly 48 hours of that approval, China's National Healthcare Security Administration assigned the device a reimbursement billing code, completing the pathway from market authorization to clinical reimbursement in days rather than months. In June 2026, the NMPA issued classification and naming guidelines giving developers a framework for sorting BCI device types by risk class and regulatory pathway.
Together, this infrastructure covers terminology, ethics, classification, non-clinical safety standards, commercial approval pathways, reimbursement integration, and AI training data quality.
Clinical and Commercial Landscape
China's commercial deployment leads the field: Neuracle's NEO system (搜索) is the only invasive BCI commercially approved anywhere in the world. In clinical trial volume, China has matched the United States in active invasive BCI trials, with plans to enroll more than 50 patients nationwide.
The technical picture is more nuanced. The NEO system (搜索) uses fewer signal-detection channels than competitors such as Neuralink (搜索) or Precision Neuroscience (搜索) and is placed epidurally rather than in direct contact with brain tissue — a design that trades signal quality for reduced surgical risk. The device is approved for patients who retain some upper arm function, limiting its addressable population compared with systems under development in the United States that target complete paralysis.
Other Chinese BCI companies, including BrainCo (搜索), NeuroXess (搜索), and StairMed (搜索), are in various stages of clinical development, with StairMed's Alibaba-backed $72.8 million round signaling commercial investment interest. BCI-related investment in China in the first four months of 2026 alone surpassed the entire 2025 total, according to the Chinese technology investment database IT Juzi. The domestic Chinese BCI market is projected to exceed 6 billion yuan (approximately $894 million) by 2028, driven by a large aging population, integrated health insurance coverage for BCI treatments in pilot provinces, and a commercially available product pipeline that did not exist two years ago.
Data Governance Questions Alongside Technical Progress
The standard's technical rigor sits within a legal framework that raises distinct questions about neural data access. EEG data encodes not only a patient's intended movement but also characteristic brain rhythms that may reveal cognitive states, emotional patterns, and mental health conditions.
China's National Intelligence Law, adopted in 2017, states in Article 7 that "any organization and citizen shall, in accordance with the law, support, assist, and cooperate with the state intelligence work." This obligation applies to all organizations operating in China and contains no carve-out for medical device companies or sensitive health data. China's Data Security Law (2021) and Cybersecurity Law (2017) add data localization and government-access provisions for "important data" and "critical information infrastructure operators."
A Recorded Future neural data threats analysis published in August 2026 described neural data as a prime target for state-sponsored espionage, identifying it as information that can reveal mental health conditions, emotional states, and cognitive patterns even when anonymized. In April 2025, Senators Chuck Schumer, Maria Cantwell, and Ed Markey signed a letter urging the Federal Trade Commission to initiate rulemaking establishing safeguards for neural data, calling for protections extending beyond existing biometric and health data rules and setting limits on secondary uses such as AI training or behavioral profiling.
Standards Leadership and Global Implications
Medical device standards set by major regulatory bodies tend to become de facto global benchmarks, particularly when they address a gap no competing authority has addressed. If the NMPA's EEG data quality standard gains traction with international BCI developers seeking to enter the Chinese market, it may become the de facto global specification for AI-BCI training data before any Western equivalent exists.
MERICS analysis has noted that China is simultaneously pursuing domestic standard leadership and international engagement, actively participating in ISO/IEC JTC 1/SC 43 — the international standards committee for BCI — and the International Telecommunication Union.
For patients, the development of rigorous AI-BCI data standards raises the technical floor any device must clear before reaching a clinical setting. The classification logic behind requiring AI-BCI devices to meet the highest regulatory tier — Class III in China's framework — reflects caution appropriate to devices that translate brain signals directly into physical action, where failure modes are physical harms rather than software bugs. For the approximately 15 million Americans and many more patients worldwide living with spinal cord injury (搜索), stroke (搜索)-related paralysis, and ALS (搜索), Chinese regulatory leadership in this area carries a real upside, provided the technology works as intended and reaches them.