A surgeon is about to implant a device that reads brain signals. A regulator must decide whether it counts as a medical device at all, and if so, how risky it is. For most of the past decade, neither question had a settled answer in China.
That gap is now closing. Between late 2025 and 2026, China's National Medical Products Administration (NMPA) published a stack of documents that, taken together, define, classify, name, and set review expectations for brain-computer interface (脑机接口) medical devices before the market is flooded with them — a terminology standard, two guidance documents, and a still-draft technical review points paper, each aimed at a different layer of the problem.
Key takeaways
- A terminology standard, YY/T 1987—2025, took effect on 2026-01-01 and gives the first official definition of a "brain-computer interface medical device."
- Two guidance documents on classification and naming were formally released on 2026-06-30, turning drafts into usable rules.
- A technical review points paper for implantable motor-function-compensation devices remains a draft for comment (opened 2026-08-18), not yet adopted.
- Invasive devices are uniformly classed as the highest-risk Category III; non-invasive rehab devices that avoid algorithmic aid can fall to Category II.
First, agree on what the thing is
The foundation is YY/T 1987—2025, Terminology for Medical Devices Using Brain-Computer Interface Technology (采用脑机接口技术的医疗器械 术语). NMPA announced it on 2025-09-15 (Announcement No. 91 of 2025) and it became effective on 2026-01-01. A "recommended" (recommended-industry, YY/T) standard rather than a mandatory one, it still matters because every later document references it.
Its definition is deliberately narrow. A brain-computer interface (脑机接口) medical device, per the standard, must (1) measure signals from the central nervous system — the brain and spinal cord; (2) decode those signals in real time and use the decoded command to achieve the product's main purpose, such as controlling an external device or triggering stimulation; and (3) achieve real-time, two-way interaction or closed-loop feedback between the patient's central nervous system and an external device. A plain EEG machine, or a deep-brain stimulator that only stimulates without a closed loop, does not qualify. That line in the sand is what makes the rest of the rulebook work.
Then decide how risky it is
On 2026-06-30, NMPA released two guidance documents: the Guiding Principles for the Classification of Medical Devices Using Brain-Computer Interface Technology (脑机接口医疗器械产品分类界定指导原则) and the Guiding Principles for the Generic Naming of Medical Devices Using Brain-Computer Interface Technology (脑机接口医疗器械通用名称命名指导原则). Both had earlier circulated as drafts for public comment in spring 2026; the June release moved them from consultation to usable guidance.
The classification document is where risk tiers get assigned:
- Invasive brain-computer interface (脑机接口) medical devices are managed as Category III, with classification code 12-00.
- Non-invasive devices used to treat disease are Category III (code 09-00); used for functional compensation, also Category III (code 19-00).
- Non-invasive devices used for limb-movement rehabilitation in stroke patients, without algorithmic aid, drop to Category II (code 19-00).
- Other rehabilitation products stay at Category III.
- Products aimed at healthy users — attention training, sleep regulation, gaming, smart-home control — are explicitly not medical devices at all.
That last bullet is the most consequential for the consumer market. It tells companies that a headband sold to "improve focus" is outside the medical-device system, while the same hardware pitched as a therapy lands in the strictest category.
The hard part: how to review an implant
The draft that is still moving is the Technical Review Points for Implantable Brain-Computer Interface Motor-Function-Compensation Medical Devices (植入式脑机接口运动功能代偿类医疗器械技术审评要点). NMPA's Center for Medical Device Evaluation (CMDE) opened it for public comment on 2026-08-18, with comments due 2026-09-07. As of this writing it had not been formally adopted, so it sets expectations rather than binding rules.
What it expects is revealing. The document applies to implantable devices that compensate for lost motor function after central-nervous-system injury — think a paralyzed patient moving a cursor or a robotic hand by intention. Review focuses on three pillars: signal acquisition (quality, real-time behavior, stability, reliability), signal decoding (generalization, accuracy, real-time behavior, stability), and the actuator (accuracy, real-time behavior, stability, usability). Software inside these devices is rated at the highest "severe" safety level, and applicants must supply threat-modeling and penetration-testing evidence alongside the usual algorithm-performance studies.
In other words, the regulator is treating the decoding software as a safety-critical component, not a feature.
Why this matters now
China's CMDE had already, in 2025, stood up a dedicated brain-computer interface working group to close the gap between fast-moving hardware and a slow-moving evaluation system. Companion test-method standards covering signal and energy transmission, invasive-device reliability, flexible-electrode electrical performance, and neural-signal acquisition and decoding are either issued or moving through the planning pipeline, suggesting the scaffolding will keep growing rather than stall at definitions.
For device makers, the practical message is simple: build to the definition, plan for Category III if you are invasive, and treat your decoding algorithm as a regulated medical function. For patients and clinicians, the value is predictability — a clearer path from lab bench to bedside, and a harder road for vague "neuro" claims that were never medical in the first place.
Honest limitations
This article relies on NMPA and CMDE publications, the national standards platform (samr.gov.cn), and industry reporting; it does not include confidential draft commentary or company-specific submissions. The timeline of the CMDE working group's formation comes from regulator summaries and media quoting center staff, not a single primary filing, so it is stated as context rather than a dated milestone. The implantable review points remain a draft for comment and could change before adoption; any device sponsor should read the final text. Classification codes and categories reflect the released guidance but are subject to later revision as companion standards mature. No clinical outcomes, pricing, or reimbursement figures are discussed here because the documents are about regulation, not efficacy.
What readers can do now
- If you build or fund neurotechnology, map your product against the YY/T 1987—2025 definition first — if it lacks real-time decoding and closed-loop feedback, the "brain-computer interface (脑机接口)" label may not apply and neither will the medical-device pathway.
- If you are evaluating an invasive or non-invasive neuro device for procurement, check its notified classification category (III vs II) and registration status on the NMPA database before trusting marketing claims.
- If you follow the sector, watch for the formal adoption of the implantable technical review points and the companion test-method standards — those are the next documents that will shape who can ship and who cannot.