Brain-Computer Interface Startups in Human Trials, and How Far Each Has Really Got

Every few weeks a brain-computer interface company announces a "first". It might be the first wireless implant, the first patient to post online by thought, or the first decoded sentence. Each is usually true in a narrow sense, and each is usually reported as if the technology had moved a step closer to the clinic. Sometimes it has. Often the announcement describes a rung of the ladder the company has already stood on, or one that matters far less to patients than the headline suggests.

This piece maps the main implantable BCI companies against a single evidence ladder. The ladder runs from laboratory work to routine clinical use. For each company it records what has been independently shown, what the company reports about itself, and what has been authorised by a regulator. The picture as of October 2026 is more uneven, and in places more surprising, than the funding league tables suggest. The first invasive BCI cleared for commercial sale anywhere was not built by the best-funded company. Nor was it the one with the most electrodes. It has eight channels and sits outside the brain's outer membrane.

The ladder every implant has to climb

The terms in BCI announcements have specific meanings, and conflating them causes most of the confusion.

Laboratory and animal work establishes that a device can be built, implanted and record usable signals, and gives early evidence of biocompatibility. It says nothing reliable about function in people.

Temporary human recording places a device in a patient for minutes, hours or days, usually during surgery that is happening anyway, such as epilepsy mapping or tumour resection. It shows that the device can be placed safely and records human neural signals. It does not show that the device can live in the body for years, or that a patient can use it at home.

Early feasibility studies are the first chronic implants. In the United States they run under an Investigational Device Exemption (IDE) from the Food and Drug Administration. They typically involve a handful of participants and are designed mainly to assess safety and gather preliminary evidence of function. A participant using a device daily under an IDE is taking part in research, not receiving a treatment.

Pivotal trials are larger studies designed to produce the evidence a regulator needs to authorise a device for a defined use, with pre-specified endpoints.

Regulatory authorisation permits marketing for a specific indication. In the US, depending on risk and on whether a comparable device already exists, that means 510(k) clearance, De Novo classification or premarket approval (PMA). In China it means registration by the National Medical Products Administration (NMPA), with Class III the highest-risk category. The authorisation applies to the specific claim made, not to the technology in general.

Commercial availability and routine clinical use come later still, and depend on manufacturing, surgeon training, reimbursement and health-system adoption.

Funding rounds and Breakthrough Device designations sit outside this ladder. A Breakthrough designation gives a company more interaction with the FDA and priority review. It is not evidence that a device works.

Where each company stands

The comparison below draws on regulatory announcements, trial registrations, peer-reviewed publications and, where clearly labelled, company statements. Participant numbers are the latest available figures and change frequently.

html

<table>
  <thead>
    <tr>
      <th>Company</th>
      <th>Device and placement</th>
      <th>Electrodes or channels</th>
      <th>Highest rung reached</th>
      <th>Chronic human participants</th>
      <th>Peer-reviewed human results</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Neuracle Technology (China)</td>
      <td>NEO; two implants on the dura above sensorimotor cortex; wirelessly powered, no internal battery</td>
      <td>8 physical channels</td>
      <td>NMPA Class III market approval (13 March 2026); first commercial implant July 2026</td>
      <td>36 in the feasibility and confirmatory programme, October 2023 to early 2026</td>
      <td>Trial results reported; registration programme completed</td>
    </tr>
    <tr>
      <td>Precision Neuroscience (US)</td>
      <td>Layer 7 thin-film array on the brain's surface</td>
      <td>1,024 per array</td>
      <td>FDA 510(k) clearance (April 2025) for implantation of up to 30 days</td>
      <td>None reported as permanent; company reports more than 68 patients tested, mostly intraoperatively</td>
      <td>Yes, small pilot studies (5 and 4 patients)</td>
    </tr>
    <tr>
      <td>Synchron (US/Australia)</td>
      <td>Stentrode; delivered through the jugular vein into a vessel over motor cortex</td>
      <td>16</td>
      <td>Early feasibility complete; pivotal trial being prepared</td>
      <td>4 in the Australian SWITCH study and 6 in the US COMMAND study</td>
      <td>Yes, SWITCH safety results (4 patients, 12 months)</td>
    </tr>
    <tr>
      <td>Neuralink (US)</td>
      <td>N1; flexible threads inserted into cortex by robot; fully implanted</td>
      <td>1,024</td>
      <td>Early feasibility studies in the US, Canada, the UK and elsewhere</td>
      <td>21 worldwide (company figure, January 2026)</td>
      <td>Paper on first three participants submitted October 2025; not found in print at time of writing</td>
    </tr>
    <tr>
      <td>Paradromics (US)</td>
      <td>Connexus; penetrating microelectrodes, chest transceiver, fully implanted</td>
      <td>421</td>
      <td>Early feasibility study under FDA IDE (approved November 2025)</td>
      <td>First Connect-One participant implanted 2026; study initially approved for 2</td>
      <td>None yet for the chronic study</td>
    </tr>
    <tr>
      <td>Blackrock Neurotech and academic BrainGate groups (US)</td>
      <td>Utah array; penetrating microelectrodes through a skull connector</td>
      <td>96 per array; up to 256 in recent speech studies</td>
      <td>Long-running academic feasibility research</td>
      <td>Dozens of people over two decades, across academic studies</td>
      <td>Yes, the field's largest body of published human evidence</td>
    </tr>
  </tbody>
</table>

Neuracle: the narrowest claim got there first

On 13 March 2026, China's NMPA granted Class III market approval to NEO, developed by Neuracle Technology. That made it the first invasive brain-computer interface authorised for commercial use anywhere. The approved indication is narrow: adults with tetraplegia from cervical spinal cord injury. The system decodes the intention to grasp and drives a pneumatic glove that closes the hand.

The device is less invasive than most of its American competitors. Two coin-sized units sit on the dura mater, the brain's tough outer membrane, through small openings in the skull. They do not penetrate brain tissue. The system has eight physical recording channels, which Neuracle says its decoding software expands into a larger set of virtual channels. It is powered wirelessly through an external coil and has no internal battery.

The approval rested on a multicentre programme that ran from October 2023 to early 2026 with 36 participants, 4 in feasibility and 32 in confirmatory studies. It was conducted at Xuanwu Hospital and Beijing Tiantan Hospital in Beijing and Huashan Hospital in Shanghai. The first commercial implant was performed at Huashan Hospital on 13 July 2026. Shanghai's supplementary insurance scheme has added partial coverage for the device's consumables.

Two qualifications matter. Neuracle says it is aiming for an operational life of more than 20 years, but the longest chronic follow-up reported so far is under three years. And an epidural system with eight channels captures far coarser signals than an intracortical array. That is appropriate for a grasp command, but not obviously for decoding speech or fine cursor control. NEO's approval shows what is achievable when a company matches a modest signal to a narrowly defined clinical task, not that the technology has matched its intracortical rivals.

China's other programmes are earlier on the ladder. Chinese state media reported in July 2026 that a formal clinical trial of NeuCyber's semi-invasive Beinao-1 system had launched in March, with enrolment under way.

Precision Neuroscience: cleared, but for 30 days

Precision Neuroscience, founded in 2021, received FDA 510(k) clearance for its Layer 7 Cortical Interface in April 2025. The company describes this as the first full regulatory clearance for a next-generation BCI company. The description is accurate, but the clearance is easy to overread.

The authorisation covers implantation for up to 30 days, for recording, monitoring and stimulating cortical activity. In practice that is the setting of neurosurgical mapping, where surgeons need to know which areas of cortex to spare. It does not authorise a permanent implant or a home-use communication device. Layer 7 is a thin, flexible film with 1,024 electrodes that rests on the brain's surface without penetrating it. Several arrays can be placed side by side; in one 2024 procedure, four arrays put 4,096 electrodes on the cortex at once.

The company says Layer 7 had been tested in more than 68 patients by January 2026, a large number for the field. Most of those recordings were temporary, during surgery. The peer-reviewed human evidence is small. One 2025 study combined animal work with pilot testing in five patients. A 2026 study from Johns Hopkins in four awake-craniotomy patients reported 77.5% accuracy on a four-word speech classification task and 78–84% accuracy in cursor control. Those figures come from short sessions in an operating theatre, not from daily use.

Precision's strategy is deliberate. A temporary surgical device can reach the market, generate revenue and build a large dataset of human cortical recordings, while the company develops a fully implantable wireless version. The company describes that version as a near-term goal but has not published a trial timeline. The clearance is a commercial and regulatory foothold, not evidence that a permanent Layer 7 implant would be safe or effective.

Synchron: no open-brain surgery, a pivotal trial pending

Synchron's Stentrode takes a different route to the brain, through the blood vessels. A stent-like scaffold carrying 16 electrodes is threaded up from the jugular vein and deployed in the superior sagittal sinus, a large vein running along the top of the brain next to the motor cortex. No craniotomy is needed. The recordings are coarser than those of penetrating arrays, but enough for users to select items on a screen and issue clicks.

Synchron has the most mature chronic safety record among the US startups. Its Australian SWITCH study, published in JAMA Neurology in January 2023, followed four patients with severe paralysis for 12 months. It reported no device-related serious adverse events, no blood clots and no migration of the device. The US COMMAND early feasibility study implanted six patients. Synchron announced in September 2024 that all six met the primary safety endpoint over 12 months, with no device-related serious adverse events involving the brain or blood vessels. The Stentrode received FDA Breakthrough Device designation in 2020.

The next step is a pivotal trial, the stage at which a company must show, with pre-specified endpoints, that the device benefits patients enough to justify authorisation. Synchron raised a $200 million Series D in November 2025 to fund that trial and prepare for commercial launch. As of October 2026 the pivotal study has been announced as a plan, not reported as enrolling. The Stentrode is not approved by the FDA or any other regulator. The company's lead in chronic safety data is real. Whether a 16-electrode device can show enough functional benefit to satisfy regulators is the question its pivotal trial will have to answer.

Neuralink: the most participants, the least published

Neuralink, founded in 2016, has the highest profile and has raised the most money in the field. That includes a $650 million Series E in June 2025, which was widely reported to value the company at about $9 billion. Its N1 implant places 1,024 electrodes on flexible threads into the cortex using a surgical robot. The device is fully implanted, with no wire through the skin, and communicates wirelessly.

The first human implant took place in January 2024 under an FDA IDE. Neuralink has since opened studies in Canada, the UK and elsewhere. University College London Hospitals reported that seven participants had been implanted in its GB-PRIME study between October and December 2025. In January 2026 the company told Reuters that 21 participants were enrolled worldwide, up from 12 the previous September. It says there have been no serious device-related adverse events. The FDA has granted Breakthrough Device designation to Neuralink's Blindsight vision project, which has not yet been implanted in people.

By participant numbers, Neuralink has moved fast for an early feasibility programme. The constraint is the evidence base. Almost everything known about how the N1 performs comes from company presentations, livestreams and participants' own accounts. In October 2025 Neuralink said it had submitted a paper on its first three participants, including safety outcomes, to the New England Journal of Medicine. At the time of writing we had not found that paper in print. The company has also publicly acknowledged a problem in its first participant: some threads retracted from the cortex in the weeks after implantation, reducing the number of working electrodes. Neuralink said it compensated with software changes.

None of this means the device performs worse than reported. It means the claims cannot yet be checked in the way claims from academic BCI groups routinely are. Every Neuralink participant is in an early feasibility study. Nothing the company has done so far amounts to a regulatory authorisation or an approved treatment.

Paradromics: designed for data rate, now in people

Paradromics, based in Austin, Texas, built its Connexus system around a single priority: capturing as much neural information as possible for speech decoding. The implant carries 421 penetrating microelectrodes made of platinum-iridium, which record from individual neurons. A transceiver implanted in the chest sends data wirelessly through the skin. The company has cited a preprint reporting more than 200 bits per second of information transfer in animal models. That figure comes from preclinical work and is not a human result.

Paradromics' route into people followed the ladder closely. In 2025 a Connexus device was placed temporarily in a patient undergoing epilepsy surgery at the University of Michigan, confirming that it could be implanted safely and record human signals. In November 2025 the FDA approved an IDE for the Connect-One early feasibility study. The study runs at UC Davis, Massachusetts General Hospital and the University of Michigan and was initially approved for two participants with severe motor and speech impairment. The first chronic implant, in a woman with motor neuron disease, was announced in June 2026. In September 2026 the company reported that she had used the system for real-time speech, including a live phone call with her grandchildren. It said the decoded output matched every intended word in the exchanges it described.

That is a striking report. At this stage it is a company press release about one participant, without published accuracy rates, vocabulary size or words-per-minute figures. The academic speech-decoding results Paradromics is implicitly measured against were published with all of those details.

The benchmark: two decades of academic evidence

Most of the startups are, in effect, trying to turn academic results into products. Those results come overwhelmingly from the Utah array, a bed-of-nails electrode made by Blackrock Neurotech, used for two decades by academic consortia such as BrainGate.

The Utah array is not elegant by current standards. It has 96 electrodes and connects through a pedestal fixed to the skull, which limits home use and carries an infection risk. But the evidence produced with it is the most extensive in the field, and it is published in detail. The clearest recent example appeared in the New England Journal of Medicine in August 2024. A UC Davis team implanted four Utah arrays, 256 electrodes in total, in a 45-year-old man with ALS. After about 30 minutes of calibration, the system decoded his attempted speech with 99.6% word accuracy on a 50-word vocabulary. On a vocabulary of 125,000 words, accuracy reached 97.5% after further data collection. Those figures come with participant details, calibration times, error rates and methods that other groups can scrutinise.

Blackrock itself has had a turbulent commercial history, including a $200 million investment from Tether in 2024. Its role in this comparison is less as a startup racing to market than as the source of the evidence base. The academic standard is detailed publication, small but well-documented cohorts and long follow-up. It is the measure against which newer companies' claims should be read.

What the comparison shows

Set side by side, these programmes reveal patterns that individual announcements obscure.

Electrode count has not predicted regulatory progress. The first commercially approved invasive BCI has eight channels. The first FDA clearance for a new-generation BCI company went to a device approved only for 30-day use. The highest-channel-count chronic implants, from Neuralink and Paradromics, remain in early feasibility. More channels allow richer decoding, which matters for speech, but regulators authorise specific claims. The quickest route has gone to companies that made narrow claims that could be tested with modest signals.

Less invasive has moved faster. Neuracle's device sits outside the dura, Precision's on the brain's surface, and Synchron's inside a blood vessel. All three have reached a regulatory milestone or completed early feasibility. The penetrating arrays offer higher-quality signals, but must also show that implants inside brain tissue stay safe and keep working over years. That is a harder and slower case to make.

Funding and published evidence are only loosely related. The best-funded company has the most chronic participants and the thinnest peer-reviewed record. Companies with far less money have published human data or completed formal registration programmes. Investors are pricing in eventual markets; regulators and clinicians want evidence that has been checked.

The human numbers are still very small. Neuracle's registration programme involved 36 participants. Every other chronic programme counts its participants in single or low double figures. The entire field's experience of chronic implants is smaller than a single mid-sized trial in most areas of medicine. Long-term failure modes, such as electrode degradation, tissue response and device breakdown, will only become clear as follow-up extends.

Temporary and chronic results are different kinds of evidence. A device that records well for an hour in an operating theatre has not shown that it will work for years in someone's home. Several companies report large numbers of temporary placements alongside small numbers of chronic implants. Those figures should never be added together.

html

<table>
  <thead>
    <tr>
      <th>Milestone in an announcement</th>
      <th>What it establishes</th>
      <th>What it does not establish</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Animal results or preclinical data rate</td>
      <td>The device can be built and record signals in another species</td>
      <td>Safety, signal quality or usefulness in people</td>
    </tr>
    <tr>
      <td>Temporary human implant during surgery</td>
      <td>The device can be placed in a person and record human neural signals</td>
      <td>Long-term safety, durability or home use</td>
    </tr>
    <tr>
      <td>FDA Breakthrough Device designation</td>
      <td>The FDA agrees the device may address an unmet need and will offer more interaction</td>
      <td>That the device works or will be approved</td>
    </tr>
    <tr>
      <td>IDE approval or first chronic implant</td>
      <td>A regulator has allowed a small study to start</td>
      <td>Efficacy, or anything beyond preliminary safety</td>
    </tr>
    <tr>
      <td>Single-participant demonstration</td>
      <td>The system can achieve a function in at least one person</td>
      <td>How often it works, for whom, or how reliably</td>
    </tr>
    <tr>
      <td>Peer-reviewed publication</td>
      <td>Methods and results have been examined by independent experts</td>
      <td>Generalisation to larger or different populations</td>
    </tr>
    <tr>
      <td>Market authorisation</td>
      <td>A regulator accepts the evidence for a specific indication and duration</td>
      <td>Usefulness for other conditions or uses beyond the approved claim</td>
    </tr>
  </tbody>
</table>

How to read the next announcement

The next round of BCI news will almost certainly include Synchron's pivotal trial, further Neuralink participant counts and the first published Connect-One data. More Chinese approvals may follow, possibly a first US authorisation for a chronic implant, and eventually long-term follow-up from the earliest participants.

A few questions sort the substance from the noise. Is the result from a temporary placement or a chronic implant? How many participants does it describe? Has the result been published with methods and error rates, or only presented by the company? What exactly has a regulator authorised, for which patients and for how long? And does the claim describe a function that has been shown in people, or one the company plans to develop?

The field has made real progress. A man with ALS can speak through a decoder with 97.5% accuracy. A woman with motor neuron disease can hold a phone conversation through a fully implanted device. A man with tetraplegia in Shanghai has received a brain implant as a commercially authorised medical device. None of these is yet routine clinical care, and the distance between each company's first rung and the last remains larger than most announcements imply.