health
Clinically useful brain implants become established by 2126.
Last revised Sep 11, 2026.
Current estimate
probability this has happened by each date
The full test below is met by 31 December 2035. The approval leg is already half-satisfied, so this horizon turns almost entirely on whether the worldwide implanted base reaches five figures inside nine years from a base of roughly one hundred.
The full test below is met by 31 December 2050. Twenty-four years is long enough for a second and third jurisdiction to approve and for indications to widen beyond tetraplegia, but the 10,000-recipient leg still requires surgical delivery capacity that does not exist today.
The full test below is met by 31 December 2100. At this horizon the dominant risk is no longer that the technology fails but that it is bypassed, by biological repair of the injury or by non-invasive decoding good enough to remove the reason to open a skull.
An implanted device that acquires signals from inside the skull - intracortical, subdural or epidural - and DECODES them into commands that restore or substitute a lost motor, sensory or communication function is, simultaneously: (a) authorised for marketing by at least two of the FDA, the EU MDR system, China's NMPA and Japan's PMDA; (b) carried by at least 10,000 living recipients worldwide; and (c) reimbursed by a national health system or a major insurer in at least one of those jurisdictions. Cochlear implants, retinal prostheses, peripheral nerve devices and open-loop deep brain stimulation are excluded, because they stimulate or suppress without decoding neural intent.
Resolution criteria
- 2035
- The full test below is met by 31 December 2035. The approval leg is already half-satisfied, so this horizon turns almost entirely on whether the worldwide implanted base reaches five figures inside nine years from a base of roughly one hundred.
- 2050
- The full test below is met by 31 December 2050. Twenty-four years is long enough for a second and third jurisdiction to approve and for indications to widen beyond tetraplegia, but the 10,000-recipient leg still requires surgical delivery capacity that does not exist today.
- 2100
- The full test below is met by 31 December 2100. At this horizon the dominant risk is no longer that the technology fails but that it is bypassed, by biological repair of the injury or by non-invasive decoding good enough to remove the reason to open a skull.
- 2126
- An implanted device that acquires signals from inside the skull - intracortical, subdural or epidural - and DECODES them into commands that restore or substitute a lost motor, sensory or communication function is, simultaneously: (a) authorised for marketing by at least two of the FDA, the EU MDR system, China's NMPA and Japan's PMDA; (b) carried by at least 10,000 living recipients worldwide; and (c) reimbursed by a national health system or a major insurer in at least one of those jurisdictions. Cochlear implants, retinal prostheses, peripheral nerve devices and open-loop deep brain stimulation are excluded, because they stimulate or suppress without decoding neural intent.
Probability history
| Revised | State | p(2035) | p(2050) | p(2100) |
|---|---|---|---|---|
hf-forecast-0196 | researched | 15% no prior revision | 60% no prior revision | 80% no prior revision |
hf-forecast-0003 | initial prior | — | — | — |
Strongest evidence for
- The world's first approved implanted BCI exists: China's NMPA authorised the NEO epidural device in March 2026 for tetraplegia after a 32-patient multicentre registration trial, and the device was assigned a national health-insurance code within days, which satisfies a meaningful part of both the approval and the reimbursement legs in one jurisdiction (hf-evidence-0061).
- The capability is demonstrated in sustained, unsupervised, real-world use rather than only in the laboratory: one participant used a speech implant at home for more than 3,800 hours over nearly two years, producing 1,960,163 words at above 99 percent decoded word accuracy (hf-evidence-0066).
- Chronic hardware durability is better than the field's own reputation suggests: across 14 participants and up to 7.6 years, intracortical arrays recorded spiking waveforms on 35.6 percent of electrodes with only a 7 percent decline over the enrollment period, and no BrainGate safety event ever required explantation (hf-evidence-0067).
- A regulatory route exists in three jurisdictions at once: an NMPA approval, an FDA final guidance specific to implanted BCIs since May 2021, and a first US 510(k) clearance for a cortical interface in March 2025 (hf-evidence-0061, hf-evidence-0069).
- The precedent base is favourable: three classes of implanted neural device already hold approvals, one of them for four decades, so neither the surgery, the regulator, nor the payer is being asked to do something unprecedented in kind (hf-evidence-0064).
- The horizon is a century, the eligible population is large and growing with population ageing, and abundant evidence that something has not happened yet is weak evidence that it will not happen within a hundred years (hf-day-0004-takeaway-02).
Strongest evidence against
- Approval does not produce adoption. The closest analogue, approved 42 years ago, insurance-covered and exceeding a million recipients, still reaches only 5.6 to 12.7 percent of the people eligible for it, and 0.33 percent of the broader hearing-loss population in a 6.92-million-patient claims dataset (hf-evidence-0063).
- No pivotal trial for an implanted BCI is registered in the United States, and the flagship intracortical programme, BrainGate2, has been running since May 2009, is still titled a feasibility study, still lists enrollment as estimated rather than actual, and carries an estimated primary completion date of September 2038 (hf-evidence-0065).
- Implanted neurotechnology has a demonstrated pattern of leaving its recipients stranded: Argus II with more than 350 recipients, Nuvectra with at least 3,000, and Autonomic Technologies with more than 700, all abandoned by companies that failed rather than by a technology that did (hf-evidence-0068).
- The one device actually approved is the least capable one in the field. NEO is epidural and drives a grasp assist; it is not the speech or cursor decoding that the headline results come from, so the approval and the capability are not yet the same object (hf-evidence-0061).
- The first US clearance is for a device that must be removed within 30 days, which is the opposite shape from a chronic implant (hf-evidence-0069).
- Even at best, roughly two thirds of the electrodes on a chronically implanted array are not recording spiking units, which bounds what a given surgery buys (hf-evidence-0067).
- Delivery capacity is the unexamined term. Every implant requires a neurosurgeon, an operating theatre and lifelong device support, and no source consulted in this cycle gives a count of the functional-neurosurgery workforce that would have to perform the procedures.
Present uncertainties
- Whether a surgical and support system capable of tens of thousands of implants is ever built, and who pays for it. Almost all the movement between the 2035 and 2050 midpoints sits on that single term, and it is a question about health-system capacity and reimbursement rather than about neuroscience.
Next discriminating observation
The cumulative NEO implant count in China twelve months after approval, against the 36 implanted during its trials. That number is a direct measurement of the approval-to-adoption term this forecast now rests on, it will exist, and no other available observation isolates that term as cleanly. Secondarily: the first registered PIVOTAL implanted-BCI trial in the US or EU, which would date the second-jurisdiction approval leg.
Sources
- The world's first approved implanted brain-computer interface was authorised in March 2026, in China, and given an insurance code days later
China's National Medical Products Administration approved NEO, an epidural brain-computer interface developed by Neuracle Technology with Tsinghua University, in March 2026, for 'patients between 18 and 60 who have paralysis in all limbs due to spinal cord injuries but still have some residual function in their arms'. The sensors 'were placed on his dura mater, the tough outer layer of tissue that covers and protects the brain'. MIT Technology Review reports that 'Days after NEO was approved, China started incorporating it into the country's health insurance system by assigning it a unique code.' The supporting registration trial is ClinicalTrials.gov NCT06990412, sponsor Neuracle Medical Technology (Shanghai) Co., Ltd., enrollment 32 actual, status Completed, start 28 May 2025, completion 10 January 2026.
Limit: NEO is epidural, not intracortical: it reads field potentials from above the dura and drives a grasp assist, which is a narrower capability than the speech and cursor decoding that produces the field's headline results. The approval is in one jurisdiction, under one regulator, for one narrow indication. It establishes that the regulatory path exists; it does not establish that the device is widely used.
The registry record was fetched directly from the ClinicalTrials.gov API v2 on 2026-09-11 and its fields are quoted above. The approval itself is corroborated independently by a 60-author landscape analysis (arXiv 2607.07185, submitted 8 July 2026), which states that as of June 2026 China had approved 'the world's first approved semi-invasive implantable BCI'. The widely reported approval date of 13 March 2026 is consistent across sources but was NOT confirmed against an NMPA primary document, which could not be fetched.
- China's BCI pipeline is 134 registered trials, and 79.1 percent of it is non-invasive
A landscape analysis by Chen, Qing, Lin and 57 co-authors, submitted 8 July 2026, states: 'We analyzed 134 clinical trials from the Chinese Clinical Trial Registry (ChiCTR), 26 IITs, and five BCI-related products approved by the National Medical Products Administration as of June 2026.' It reports that 'Non-invasive systems predominated, accounting for 79.1% of registered studies, with stroke rehabilitation as the leading indication (65.0%)', and that the five approvals comprise 'the world's first approved semi-invasive implantable BCI, an invasive closed-loop deep brain stimulation system with real-time local field potential recording, and three non-invasive EEG-based rehabilitation systems.' The authors name as remaining barriers 'long-term implant stability, standardization of clinical infrastructure and workflows, and generalizability of decoding algorithms'.
Limit: A preprint, not peer-reviewed, and restricted to China. Only one of the five approved products is the implanted decoding BCI this forecast concerns, so the headline 'five approvals' overstates the relevant count fourfold if read carelessly.
The arXiv abstract page was fetched directly on 2026-09-11 and the sentences above are quoted from it. The PDF could not be text-extracted, so only the abstract is quoted; no figure or table in the body was read.
- A neural implant approved 42 years ago still reaches under 13 percent of the people eligible for it
A 2024 JAMA Otolaryngology-Head & Neck Surgery analysis states that for cochlear implants 'utilization rates remain low in the US (5.6%-12.7%)', and its own claims analysis found that 'CI utilization overall was 0.33% of all patients with SNHL' across 6.92 million patients with bilateral sensorineural hearing loss in PearlDiver Mariner data covering 2010-2022. NIDCD states that 'As of December 2019, approximately 736,900 registered devices have been implanted worldwide. In the United States, roughly 118,100 devices have been implanted in adults and 65,000 in children', and elsewhere that more than one million had been implanted worldwide as of July 2022. Cochlear implants have been FDA-approved for adults since 1984.
Limit: The 0.33 percent figure uses all sensorineural hearing loss as its denominator, and most of that population are not implant candidates, so the 5.6-12.7 percent candidate-restricted range is the one that bears on this forecast. Cochlear implant under-utilization is partly driven by referral and awareness failures specific to audiology, which a high-visibility BCI might not inherit. The figure bounds how far approval alone carries a neural implant; it does not fix the ceiling for a different device in a different specialty.
PMC10884948 and the NIDCD cochlear implants page were both fetched directly on 2026-09-11 and the sentences above are quoted from those fetches. The 5.6-12.7 percent range is stated by that paper as a summary of prior studies rather than as its own measurement; the 0.33 percent figure is the paper's own result.
- Three implanted neural devices were already approved decades ago, and one of them restores a major sensory function
FDA PMA P960009, the Medtronic Activa Tremor Control System, carries decision date 07/31/1997 and is indicated for 'UNILATERAL THALAMIC STIMULATION FOR THE SUPPRESSION OF TREMOR IN THE UPPER EXTREMITY IN PATIENTS WHO ARE DIAGNOSED WITH ESSENTIAL TREMOR OR PARKINSONIAN TREMOR NOT ADEQUATELY CONTROLLED BY MEDICATIONS'. The NeuroPace RNS System was approved under PMA P100026 on 11/14/2013. The Argus II Retinal Prosthesis System was authorised under Humanitarian Device Exemption H110002 on 02/13/2013, and the applicant of record on FDA's live database is now 'Cortigent, Inc.' rather than Second Sight. Cochlear implants exceed one million recipients worldwide.
Limit: None of these decodes neural signals into commands. Cochlear implants stimulate the auditory nerve from an external microphone, deep brain stimulation is open-loop stimulation that suppresses a symptom, and Argus II is a retinal rather than a brain device. They establish what 'approved implant restoring a major function' already means, which is why they force the resolution criterion to be rewritten; they are not evidence that a decoding BCI will travel the same path.
The P960009 PMA record was fetched directly on 2026-09-11 and its decision date and indication text are quoted above. The H110002 and P100026 records were fetched via a delegated worker from the same FDA databases and are quoted from those fetches. The 1984 first cochlear implant approval could not be located in the FDA PMA database and rests on secondary sources only.
- Seventeen years in, the flagship intracortical BCI trial is still a feasibility study, and no implanted BCI pivotal trial is registered in the United States
BrainGate2 (NCT00912041) is titled a 'Feasibility Study of an Intracortical Neural Interface System for Persons With Tetraplegia', started May 2009, is still RECRUITING, lists enrollment of 27 as ESTIMATED rather than actual, and gives an estimated primary completion date of September 2038. Synchron's COMMAND study (NCT05035823) is explicitly 'An Early Feasibility Study (EFS)' with 6 participants actual. Paradromics' Connect-One (NCT07357428) is an 'Early Feasibility Study' with 2 estimated. Neuralink's PRIME (NCT06429735) is 'An Early Feasibility Study' with 15 estimated. A search for a registered pivotal implanted-BCI trial returned no such study; the one registered trial with 'Pivotal' in its title, NCT05147792, was confirmed by direct fetch to be an unrelated cardiac left-atrial-appendage device trial.
Limit: ClinicalTrials.gov is a US-centric registry, and the pivotal/feasibility distinction is an FDA construct. China's NEO reached approval through a 32-patient multicentre registration trial that this search classifies as neither. So the absence of a registered pivotal trial constrains the US timeline specifically, and says less about the world timeline than it appears to.
NCT00912041 and NCT05035823 were fetched directly from the ClinicalTrials.gov API v2 on 2026-09-11 and their fields are quoted above. NCT07357428 and NCT06429735 were fetched from the same API by a delegated worker. The absence of a registered pivotal trial is an absence of evidence from a bounded search of one registry, not a certified negative.
- One person used a speech implant unsupervised at home for 3,800 hours and nearly two million words
A Nature Medicine report (2026;32(7):2504-2510) on participant T15, 'a 45-year-old man with paralysis and severe dysarthria due to ALS', states that 'Over nearly 2 years, the participant used the BCI for more than 3,800 h at home' and that 'He communicated 183,060 sentences-totaling 1,960,163 words-at an average rate of 56 words per minute.' In formal testing, 'attempted speech was consistently decoded with more than 99% word accuracy (125,000 word vocabulary)'. Earlier peer-reviewed records in the same line include 62 words per minute at a 23.8% word error rate on a 125,000-word vocabulary (Willett et al., Nature, 2023).
Limit: N equals one. Sustained accuracy in a single participant says nothing about the fraction of candidates in whom a decoder calibrates successfully at all, and no source consulted in this cycle reports that fraction. It is a demonstration that the capability is real and durable in use, not a measure of how generally it works.
PMC13375540 was fetched directly by a delegated worker on 2026-09-11 and the sentences are quoted from that fetch. The Willett et al. 2023 figures were quoted from the open-access PMC mirror PMC10468393. Two further headline results, Metzger et al. (Nature 2023) and Card et al. (NEJM 2024), could not be fetched past their paywalls and are deliberately excluded from this record rather than quoted from search summaries.
- Chronic intracortical arrays hold up far better than feared, on about one third of their electrodes
A 20-year BrainGate dataset reports that 'On average, arrays successfully recorded neural spiking waveforms on 35.6% of electrodes, with only a 7% decline over the study enrollment period', across 14 participants, 20 arrays and 2,319 recording sessions, with implant durations from 296 to 2,780 days, a mean of 1,063 days, and a maximum of 7.6 years. An explant analysis of arrays removed from human cortex found '27.8% tip damage for P1 and 11.1% for P2' at 980 and 182 implanted days respectively, and that stimulated iridium-oxide tips showed '41.1%' tip damage against '5.6%' for non-stimulated tips on the same array. The BrainGate interim safety report covering 14 participants and 12,203 implant-days records '68 device-related adverse events, including 6 device-related SAEs', and that 'There were no safety events that required device explantation.'
Limit: A 35.6 percent electrode yield is evidently sufficient for the published decoding results, so this is a durability datum rather than a failure datum, and it argues against the common assumption that arrays fail quickly. What it does not answer is whether that yield holds at a revision rate and a price a health system will pay for, which is the question this forecast actually turns on and which no source consulted addresses.
PMC12236888 was fetched directly on 2026-09-11 and its figures are quoted above; it is the medRxiv preprint version (DOI 10.1101/2025.07.02.25330310) and has NOT completed peer review, which is why it is labelled mixed-tier here. PMC8688945 and PMC10074470 were fetched by a delegated worker from PubMed Central and are peer-reviewed.
- Three implanted neurotechnology companies failed and left their recipients with dead hardware inside them
An IEEE Spectrum investigation reports that Argus II recipients 'and more than 350 other blind people around the world with Second Sight's implants in their eyes, find themselves in a world in which the technology that transformed their lives is just another obsolete gadget', despite the company's 2019 written promise that 'Second Sight will be maintaining a full team of Customer Care and Vision Rehab Specialists accessible to you as we have in the past'. A former engineer states: 'We didn't really support the basic Argus after that. We didn't sell any more, we didn't make any more, we didn't have anything to do with it anymore.' A Nature feature records that Nuvectra's spinal-cord stimulator 'had been implanted in at least 3,000 people' before its 2019 bankruptcy, and that when Autonomic Technologies collapsed at the end of 2019 its closure 'left [patient] and more than 700 other people alone with a complex implanted medical device'.
Limit: Commercial failure of a device company is not evidence about the clinical value of the technology, and all three firms were small, venture-funded and operating in narrow indications where no large incumbent had an interest in acquiring the support obligation. It is, however, direct evidence about the durability of the ongoing service an implant recipient depends on, which is exactly what the word 'established' in this claim requires and which a device approval does not by itself supply.
The IEEE Spectrum investigation was fetched directly by a delegated worker on 2026-09-11. The Nature feature (Nature 2022, d41586-022-03810-5) was fetched from the author's own verbatim republication because nature.com refused the fetch, which is a weaker provenance and is recorded as such.
- The first US regulatory clearance in this field is for a device that must come out within 30 days
FDA 510(k) K242618 covers the device 'Layer 7-T' from Precision Neuroscience, Corp., with decision date 03/30/2025 and decision 'Substantially Equivalent (SESE)', under regulation number 882.1310, product code GYC, classification 'Electrode, Cortical'. The FDA record fetched states no duration of use. The company's own release describes clearance for implantation 'up to 30 days', for 'recording, monitoring, and stimulation of electrical activity on the surface of the brain' and 'intraoperative brain mapping', and claims 'It also marks the first full regulatory clearance granted to a company developing a next-generation wireless BCI.'
Limit: A 510(k) is a demonstration of equivalence to an existing predicate in an old device class, 'Electrode, Cortical', not an approval of a brain-computer interface. Read as the field's first approval it substantially overstates the event, and the company does not quite make that claim either. Its forecast-relevant content is the opposite of a milestone: the first US clearance is explicitly for a temporary device.
The FDA 510(k) record for K242618 was fetched directly on 2026-09-11 and its fields are quoted above. The 30-day limit, the intended-use wording and the 'first clearance' characterisation are from the company's own release, fetched directly on 2026-09-11, and are attributed to the company rather than to FDA precisely because the FDA record did not carry them.