Industries / Frontier science / Brain-computer interfaces
Trials and approval
This work moves a brain implant from its first tests in people to a product that a regulator allows on the market and that a health system will pay for.
6
Milestones achieved, from 2004 to 2026.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2004
BrainGate enters US trials
The US Food and Drug Administration (FDA) gave Cyberkinetics, a Brown University spin-off, an investigational device exemption (IDE), the permit that lets an unapproved device be tested in people, for a pilot trial of its BrainGate system: a sensor implanted in the brain whose signals are decoded to work assistive devices. Four people with tetraplegia enrolled through hospitals in Rhode Island, Massachusetts and Illinois. In 2007, under business pressure, Cyberkinetics turned to other devices and stopped paying for the research; grants kept it going, and in May 2009 the FDA gave a new IDE to an academic team led from Massachusetts General Hospital for the BrainGate2 trial.
Lessona trial paid for by one company can stall when that company's money runs short, and an academic sponsor is what kept this one running.
Credit
Cyberkinetics; Brown University; Massachusetts General Hospital
2021
FDA guidance for brain implants
On 20 May 2021 the FDA issued final guidance on implanted brain-computer interfaces (BCIs) for people with paralysis or amputation, after a draft in February 2019. It sets out the bench, animal and software tests the agency expects before a study in people, and what early feasibility studies (small first studies of a device still being developed) and later pivotal studies should measure. The FDA calls it a "leapfrog" guidance: its initial thinking on a technology early in development, which may change.
Credit
FDA, Office of Neurological and Physical Medicine Devices
2024
Regulators join a shared forum
Researchers, companies, clinicians and implant users signed the charter of the Implantable Brain-Computer Interface Collaborative Community (iBCI-CC) on 17 January 2024; it received a federal liaison letter in February and launched in March. Its workgroups draft shared positions on trial endpoints, payment, standards and neural data privacy, and it has published a consensus statement on which trial endpoints show that an implant is clinically useful. The FDA lists the iBCI-CC among the collaborative communities in which it takes part as a member.
Credit
iBCI-CC founding members; FDA
2025
China prices BCI procedures
On 12 March 2025 China's National Healthcare Security Administration (NHSA), which runs the state health insurance, listed BCI procedures as their own category in its national price guideline, with separate items for implanting an invasive BCI, removing one and fitting a non-invasive one. No invasive BCI was yet approved in China to bill for. Hubei province set the first local rates, and in late November 2025 a man surnamed Liu, paralysed by a spinal cord injury, received an implant at Tongji Hospital in Wuhan in what Sixth Tone reports as the first BCI procedure billed under the new system: 6,552 yuan (about USD 925), paid from research funds.
Lessona set price for the surgery is not payment for the device, and the first case was still paid for by research money.
Credit
NHSA; Tongji Hospital, Wuhan
2025
A 1,024-electrode array cleared
On 30 March 2025 the FDA cleared Precision Neuroscience's Layer 7-T, a thin polyimide film carrying 1,024 platinum electrodes that rests on the surface of the brain. It went through the 510(k) route, which allows sale of a device shown to be substantially equivalent to one already on the market; the comparison device was Ad-Tech's subdural strip and grid electrodes. The clearance covers temporary use, less than 30 days, for recording, monitoring and stimulating on the brain's surface.
Lessonthe route to market open today is as a short-term electrode for surgery and monitoring, not as a long-term implant for controlling a computer.
Credit
Precision Neuroscience; FDA (510(k) K242618)
2026
China approves an implanted BCI
On 13 March 2026 China's National Medical Products Administration (NMPA) gave a Class III registration, its highest risk class, to NEO, made by Neuracle of Shanghai with a Tsinghua University team. A coin-sized implant sits under the skull, above the brain's outer membrane, reads movement signals and drives a pneumatic glove that closes the hand. It is approved for adults aged 18 to 60 with a spinal cord injury at C2 to C6 who can still move the upper arm but cannot grasp. The evidence was 36 implant procedures followed for 18 months, with no device-related serious adverse events reported. Sixth Tone reports that officials describe it as the world's first invasive BCI cleared for commercial use; Paradromics, a US competitor, writes that the NHSA gave it an insurance code within 48 hours, but that the device itself is not yet reimbursed and that a full procedure costs an estimated 300,000 to 500,000 yuan.
Lessonapproval and a billing code do not make a device affordable while insurance will not pay for the implant itself.
Credit
Neuracle; Tsinghua University; NMPA
Watch for
A speech study is approved
Paradromics says the FDA approved an IDE for Connect-One, an early feasibility study of its Connexus implant to restore speech and computer control, and that it is the first company to receive IDE approval for speech restoration with a fully implantable BCI. The study will first enroll two people at UC Davis, Massachusetts General Hospital and the University of Michigan; the registry entry gives an actual start date of 31 March 2026 and a primary safety outcome of device-related adverse events.
Achieved whenspeech or typing results from study participants appear in a peer-reviewed journal or are reported by the trial hospitals.
Claimed by
Paradromics, November 2025
Medicare cover at approval
The Centers for Medicare & Medicaid Services (CMS) and the FDA proposed RAPID (Regulatory Alignment for Predictable and Immediate Device), under which CMS would propose a national Medicare coverage decision at about the time the FDA authorizes an eligible Breakthrough device. It aims to cut the wait from 9 to 12 months to 60 to 90 days, applies to Class III devices and some Class II devices, and does not give automatic coverage. Hall Render, a law firm, notes that earlier attempts (Parallel Review in 2016, a 2021 program rescinded the same year, and TCET in 2024) did not close the gap.
Achieved whenCMS issues a final notice and a device authorized under it receives a final national coverage decision within 90 days of FDA authorization.
Claimed by
CMS and FDA, April 2026
Open
01
One tracker for implant trials
A public, dated table of every registered clinical study of an implanted BCI worldwide, drawn from ClinicalTrials.gov, the Chinese Clinical Trial Registry (ChiCTR), the Australian New Zealand Clinical Trials Registry (ANZCTR), the ISRCTN registry and Germany's DRKS. Each row gives the registry number, sponsor, device, where it sits (in the brain, on its surface, above its membrane, or in a blood vessel), countries, start date, planned and actual enrollment, the primary endpoint as written, status, and a log of every status change, refreshed monthly. It uses study-level registry records only, never data about participants.
The briefCredit · unclaimed02
Four rulebooks, side by side
A clause-by-clause comparison of what the US, the European Union, China and Australia ask before an implanted BCI is first tested in people and before it can be sold: bench and animal tests, wireless and cybersecurity tests, study size, length of follow-up, endpoints, and the steps to payment. Every cell quotes and links the clause it relies on: the FDA's 2021 BCI guidance and IDE rules; the EU Medical Device Regulation (MDR); the NMPA's Class III rules, its 2025 BCI terminology standard (YY/T 1987-2025) and its 2026 classification guidance; China's separate research track run by the National Health Commission since 1 May 2026; and the Australian Therapeutic Goods Administration's (TGA) device rules.
The briefCredit · unclaimed03
A pooled table of harms
A machine-readable table of every adverse event reported in published studies of implanted BCIs and in registry results, coded with the free adverse-event terms of the International Medical Device Regulators Forum (IMDRF), with the device type, the total days of implant exposure and the page each number comes from. It uses only the counts that papers already publish, never records about individual participants, so that harm rates can be compared across devices.
The briefCredit · unclaimed
Open 01 · the brief
One tracker for implant trials
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, dated table of every registered clinical study of an implanted BCI worldwide, drawn from ClinicalTrials.gov, the Chinese Clinical Trial Registry (ChiCTR), the Australian New Zealand Clinical Trials Registry (ANZCTR), the ISRCTN registry and Germany's DRKS. Each row gives the registry number, sponsor, device, where it sits (in the brain, on its surface, above its membrane, or in a blood vessel), countries, start date, planned and actual enrollment, the primary endpoint as written, status, and a log of every status change, refreshed monthly. It uses study-level registry records only, never data about participants.
Why it is still open
The 2024 review by Patrick-Krueger, Burkhart and Contreras-Vidal counted 28 trials up to December 2023 and called for a registry. The International iBCI Patient Registry counts participants (87), not trials, and was last updated on 20 July 2025. bciintel's trial list draws only on ClinicalTrials.gov and lists about 27 studies, none from China. A July 2026 preprint by Long Chen and colleagues analysed 134 ChiCTR studies, most of them non-invasive, as a one-off snapshot to June 2026. No source found merges the registries or records changes over time.
Built with what exists
The ClinicalTrials.gov API (version 2, free JSON); the ChiCTR, ANZCTR, ISRCTN and DRKS search pages; Python; a scheduled GitHub Actions job for the monthly refresh; an AI model that reads registry text in Chinese and English and fills the fixed columns, with a person checking a sample. Models that read long records in two languages and fill a table are the advance that makes a monthly refresh affordable for one person.
First step
Run the ClinicalTrials.gov API query for brain-computer interface studies, keep the ones with an implanted device, and hand-label 50 rows to fix the categories.
Done looks like
Every implanted-BCI study returned by a published set of search queries in the five registries, each row linked to its record; device type checked by hand on a random 10% of rows with at least 95% agreement; six monthly refreshes in a row, each with a public change log.
Open 02 · the brief
Four rulebooks, side by side
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A clause-by-clause comparison of what the US, the European Union, China and Australia ask before an implanted BCI is first tested in people and before it can be sold: bench and animal tests, wireless and cybersecurity tests, study size, length of follow-up, endpoints, and the steps to payment. Every cell quotes and links the clause it relies on: the FDA's 2021 BCI guidance and IDE rules; the EU Medical Device Regulation (MDR); the NMPA's Class III rules, its 2025 BCI terminology standard (YY/T 1987-2025) and its 2026 classification guidance; China's separate research track run by the National Health Commission since 1 May 2026; and the Australian Therapeutic Goods Administration's (TGA) device rules.
Why it is still open
The FDA guidance covers only US submissions and calls itself initial thinking. Sidley Austin's 2024 analysis covers the EU only, and the 2026 accounts of China's rules by Conventus Law and BioSpectrum cover China only. Searches found no comparison of all four that cites the primary texts, and none that covers Australia.
Built with what exists
The public texts above (FDA guidance PDF, the MDR on EUR-Lex, NMPA notices, TGA guidance); an AI model with a long context window that reads Chinese and English legal text and drafts each cell with its quote; a spreadsheet published as a static web page.
First step
Take the section headings of the FDA guidance as the row list and find the matching EU, Chinese and Australian clauses for the first ten rows.
Done looks like
At least 30 requirements compared across the four jurisdictions, every cell linked to a clause, and a second reader re-checking a random 20% of cells with the error rate published.
Open 03 · the brief
A pooled table of harms
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A machine-readable table of every adverse event reported in published studies of implanted BCIs and in registry results, coded with the free adverse-event terms of the International Medical Device Regulators Forum (IMDRF), with the device type, the total days of implant exposure and the page each number comes from. It uses only the counts that papers already publish, never records about individual participants, so that harm rates can be compared across devices.
Why it is still open
The 2024 Nature Reviews Bioengineering review found that adverse events, explant reasons and signal quality are "rarely provided" in the literature. Bullard and colleagues' 2020 review of 48 people with Utah arrays found too few complications described to count, and used implant time as a lower bound instead. BrainGate's 2023 safety report covers one group's 14 participants. A 2024 meta-analysis by Barlatey and colleagues pooled complications outside the skull from 48 publications on deep brain stimulators, responsive stimulators, nerve stimulators and EEG recorders under the scalp (14.0% in all, infection 4.6%), but it covers no implanted BCI. No pooled table across BCI groups was found.
par.nsf.govchestekresearch.engin.umich.eduebi.ac.ukebi.ac.uk
Built with what exists
The Europe PMC API for open full texts; the results fields of ClinicalTrials.gov; IMDRF adverse-event terminology; an AI model that reads PDFs and extracts counts into a schema, with two people checking every extracted number.
First step
Enter the BrainGate interim safety report (68 device-related events over 12,203 implant-days) into the schema, then test the schema on the 12-month SWITCH report of the Stentrode.
Done looks like
Events per 1,000 implant-days for each device type, from at least 20 studies, every number traced to a page, and two independent extractors agreeing on at least 95% of numbers.