← Articles
Read in another language
Technology

Brain-Computer Interfaces: Inside the First Human Trials

Jayden

Analyzes global supply chains, industrial policy, and technology issues.

Published

Key points

  • Between 2024 and 2026 several companies moved implanted brain-computer interfaces (BCIs) into living human volunteers: Neuralink's PRIME study (first implant January 2024, later expanding to UK and UAE arms), Synchron's endovascular Stentrode, Precision Neuroscience's surface array and Paradromics' Connexus.
  • The strongest evidence is peer-reviewed and comes from academic teams, not press releases: a UC Davis study in the New England Journal of Medicine (August 14, 2024) decoded attempted speech from a 45-year-old man with ALS using 256 cortical microelectrodes, and a 2025 Nature follow-on synthesized an audible voice from raw neural activity.
  • Every human result here is small. Synchron's US COMMAND early feasibility study enrolled 6 patients and all 6 met the primary safety endpoint over 12 months; the two speech studies each had a single participant.
  • Regulatory milestones are being read as more than they are. FDA Breakthrough Device Designation is a fast lane, not approval; Precision's April 17, 2025 clearance covers an electrode array for up to 30 days. As of 2026 no fully implanted, general-purpose BCI has FDA marketing approval.
  • The law moved first on neural data: Colorado's HB 24-1058 (signed April 17, 2024, effective August 7, 2024) and California's SB 1223 (signed September 27, 2024, effective January 1, 2025) both require opt-in consent, while long-term safety and reversibility data for the devices themselves barely exist yet.

For decades, a chip that lets you move a cursor with a thought lived in science fiction. In 2024 it moved into operating rooms. Since then, several companies have placed neural implants in living human volunteers, Neuralink has expanded its study across four countries, and academic teams have published peer-reviewed results letting people with paralysis "speak" through a computer. This is a story about hardware, companies, trials, and the rules being written around them — not about the biology of any one disease, and not about artificial intelligence in general.

The temptation is to read every announcement as a leap toward reading minds. The reality is narrower and, in its own way, more remarkable: small numbers of profoundly paralyzed people are using implanted electrodes to do specific things — move a cursor, type, produce speech — under careful medical supervision. Sorting the measured results from the marketing is the whole task.

What a brain-computer interface actually is

A brain-computer interface (BCI) is a device that records electrical activity from the brain and translates it into a command for a computer or other machine. An implanted BCI usually sits on or in the motor cortex, the strip of brain that plans movement. When someone who is paralyzed attempts or imagines a movement, those neurons still fire; the implant reads the pattern, and software maps it onto an action — a cursor moving left, a letter being selected, a synthesized word.

The medical targets are specific and serious: paralysis from spinal-cord injury or stroke, and neurodegenerative conditions such as ALS (amyotrophic lateral sclerosis) that can leave a fully aware person unable to move or speak. For these patients, "control a cursor" is not a party trick; it can mean the return of independent communication. That framing matters, because it separates the genuine clinical work from the consumer "mind-reading" hype that often rides alongside it.

Implants differ mainly in how invasive they are, and that trade-off runs through the whole field. Penetrating microelectrode arrays push tiny needles into brain tissue and capture the sharpest signals from individual neurons. Surface arrays lay electrodes on top of the cortex without piercing it, trading some signal resolution for lower surgical risk. And at least one approach avoids opening the skull altogether. More signal generally means more capability but more risk; less invasive generally means safer but coarser. No design has "won."

The companies racing to put electrodes in human brains

The most famous name is Neuralink. Its PRIME study (Precise Robotically Implanted Brain-Computer Interface) places the N1 implant in the motor cortex using a surgical robot, and the company reports that participants with quadriplegia have controlled cursors, played games, and typed by intention [source: Neuralink, 2025]. Neuralink says it has moved from its first human implant in early 2024 to more than a dozen participants and has begun enrolling internationally, including UK and UAE arms registered on public trial registries [source: ClinicalTrials.gov, 2025]. These are important to label clearly: the participant counts and performance figures are company statements, not peer-reviewed clinical publications. They may well hold up, but independent, published validation of Neuralink's numbers lags behind its announcements.

Synchron takes the least invasive route. Its Stentrode is threaded through a blood vessel — up the jugular vein and into a large vein sitting against the motor cortex — so it can be placed without open-skull surgery. In the US COMMAND early feasibility study, six patients with severe upper-limb paralysis received the device; over twelve months, all six met the primary safety endpoint, with no device-related serious adverse events causing death or permanent increased disability, and the company reported the device was deployed on target in every case [source: Synchron, 2024]. That is a real, prospective safety result — but note the scale (six people) and that efficacy was described qualitatively rather than proven in a large controlled trial.

Precision Neuroscience is pursuing the surface-array path and reached a regulatory milestone first: in April 2025 the FDA cleared its Layer 7 cortical interface — a film thinner than a hair carrying 1,024 electrodes — to record and stimulate for up to 30 days [source: Precision Neuroscience, 2025]. Clearance of a limited-duration electrode array is not the same as approval of a complete, permanently implanted therapeutic system, but it is a concrete regulatory step. Paradromics placed its Connexus device in a person temporarily during an epilepsy surgery in 2025, then in June 2026 the University of Michigan performed the first implantation for an FDA-approved early feasibility study; the array carries 421 microelectrodes and transmits wirelessly to a transceiver in the chest [source: Michigan Medicine, 2026].

Behind these companies stands a quieter foundation. Blackrock Neurotech makes the Utah Array used in much of the field, and the academic BrainGate consortium — Brown University, Massachusetts General Hospital, Stanford, UC Davis and others — has produced two decades of peer-reviewed BCI research on cursor control, robotic arms, and speech. When you want to know what has actually been measured rather than announced, the academic literature is usually where to look.

The clearest wins are in speech

The strongest evidence so far is not a flashy demo; it is a pair of peer-reviewed speech studies. In 2024, a team at UC Davis, working with a 45-year-old man with ALS, used 256 cortical electrodes to decode his attempted speech with roughly 97% accuracy over a 125,000-word vocabulary, published in the New England Journal of Medicine [source: New England Journal of Medicine, 2024]. In 2025, a follow-on study in Nature went further, synthesizing an audible voice from his neural activity almost instantly — turning raw brain signals into sound within about a hundredth of a second, with closed-loop feedback so he could hear himself [source: Nature, 2025].

These results earn the word "breakthrough" more honestly than most, because they cleared peer review and reported hard numbers. But two caveats keep them grounded. First, each study involved a single participant; high performance in one person is a proof of concept, not a population result. Second, this is decoding attempted speech from someone who wants to speak — it is emphatically not reading private, unspoken thoughts. The distinction between decoding a deliberate communication attempt and "reading a mind" is the difference between the science and the hype.

Reading the numbers carefully

Because the field mixes peer-reviewed science with competitive product announcements, the same three habits that serve any technology story serve here.

Separate company claims from independent validation. A press release reporting that participants type quickly or hit a certain cursor "bit rate" is a company describing its own results; a paper in Nature or NEJM has passed outside scrutiny. Both can be true, but they carry different evidential weight, and right now the peer-reviewed evidence base is dominated by academic and hospital teams, while some of the most-quoted performance figures come straight from companies [source: Neuralink, 2025].

Respect small numbers. Almost every result in this article — six patients here, one participant there — comes from early feasibility studies whose first job is safety, not proof of benefit. Early feasibility is exactly what it sounds like: a first careful look in a handful of people. It is how responsible device development starts, and it is also why sweeping claims about what BCIs "can do" are premature.

Be careful with cause and comparison. When a participant improves at a task over weeks, some of that gain is the device and some is learning, practice, and better software tuning; performance figures are not all attributable to the hardware alone. And a number from one company's setup cannot be directly compared with another's, because the tasks, implants, and patients differ. Impressive demonstrations are real, but they are demonstrations, not head-to-head evidence.

The regulatory path, in plain terms

No fully implanted, general-purpose BCI has FDA marketing approval as of 2026. Every human implant discussed here is happening under an investigational framework, and understanding that framework explains a lot of the news.

The usual sequence runs like this. A device can receive an FDA Breakthrough Device Designation, which speeds up interaction with regulators for serious conditions — Neuralink obtained such designations, including one for a speech application in 2025, and Synchron received one back in 2020 [source: MassDevice, 2025]. Designation is not approval; it is a fast lane, not a finish line. To implant in people, a company needs an Investigational Device Exemption (IDE) to run an early feasibility study — the small, safety-first trials described above. Positive early results lead toward a larger pivotal trial; Synchron has said it is preparing one, the step it must clear before seeking a premarket approval (PMA) to sell a device [source: Synchron, 2024]. Only at the end of that road does a BCI become an approved medical product.

The pathway is also widening beyond paralysis and ALS. In 2026, CorTec's system received a Breakthrough Device Designation for stroke motor rehabilitation, reportedly the first BCI so designated for that use [source: CorTec, 2026] — a reminder that "BCI" is a category of very different devices and indications, not a single product.

The ethics: neural data, reversibility, and long-term safety

The hardest questions are not purely technical. The most novel is neural data: signals recorded directly from the brain are among the most intimate data imaginable, and the law is only beginning to respond. In 2024 Colorado became the first US state to add neural data to its privacy law, and California extended its consumer-privacy act to cover neural data effective January 2025; both generally require opt-in consent to collect and process it, and other states have introduced similar bills [source: ArentFox Schiff, 2024] [source: Arnold & Porter, 2025]. Supporters argue this pre-empts a genuine harm before consumer neurotech scales; skeptics note the definitions are new and untested, and that medical-research data is often handled under separate rules. It is an unsettled debate, not a solved problem.

Then there is reversibility and long-term safety. An implant is not a phone you upgrade; removing or replacing one is surgery, and tissue can react to a foreign body over years. Early studies emphasize safety endpoints precisely because the long-term record is still being written — the multi-year data simply does not exist yet for most of these devices. Less invasive designs, such as Synchron's endovascular approach or surface arrays, are partly a response to that worry, trading some capability for lower surgical and explantation risk.

Finally, the evidence itself is thin by design. Small trials cannot tell us how these systems perform across thousands of diverse patients, how durable the benefits are, or how often they fail. Enthusiasts see an inflection point; cautious clinicians see promising pilots that must still prove themselves at scale. Both readings are defensible, and honest coverage holds them together rather than picking the exciting one.

What to watch

Three things will show whether 2024-2026 was a true turning point or an early chapter. First, watch for peer-reviewed clinical results from the companies, not just press releases — independent publication of implant performance and durability is the signal that matters most. Second, watch the regulatory ladder: whether an early feasibility study graduates to a pivotal trial and, eventually, the first premarket approval for an implanted BCI. Third, watch the rules, from neural-data privacy laws to standards for device safety, reversibility, and informed consent, since those will shape who can build these systems and how.

The measured story is genuinely hopeful and genuinely unfinished. People who could not communicate are, in careful trials, doing so again. That is worth taking seriously — and so is the distance still to travel between a handful of implanted volunteers and a proven, approved medical technology. Watch the peer review, watch the trials, and watch what regulators actually clear, not just what gets announced.

Charts

Electrodes per array, by device

Electrodes per array, by deviceUC Davis / BrainGate (NEJM 2024) 256 electrodes, Paradromics Connexus 421 electrodes, Precision Layer 7 1,024 electrodes256 electrodesUC Davis / BrainGate (NEJM 2024)421 electrodesParadromics Connexus1,024 electrodesPrecision Layer 7
Electrode counts as reported by each producer: 256 penetrating microelectrodes across four arrays in the UC Davis speech study (New England Journal of Medicine, 2024), 421 microelectrodes in Paradromics' Connexus (Michigan Medicine / Paradromics, 2026), and 1,024 electrodes in Precision Neuroscience's Layer 7 thin-film surface array (Precision Neuroscience, 2025). These are different device classes — penetrating versus surface — so a higher count is not evidence of better performance, and the figures come from three separate producers rather than one common measurement.

Human participants in the results discussed

Human participants in the results discussedSynchron COMMAND (safety EFS) 6 participants, NEJM 2024 speech decoding 1 participants, Nature 2025 voice synthesis 1 participants6 participantsSynchron COMMAND (safety EFS)1 participantsNEJM 2024 speech decoding1 participantsNature 2025 voice synthesis
Study size, not performance. Synchron's US COMMAND early feasibility study enrolled 6 patients (Synchron, 2024); both peer-reviewed speech studies reported a single participant (New England Journal of Medicine, 2024; Nature, 2025). Neuralink is absent from this axis on purpose: its participant total is company-reported and given only as a direction — more than a dozen and expanding — rather than a published figure.

Timeline

  1. Synchron receives an FDA Breakthrough Device Designation for its endovascular BCI — a faster regulatory lane, not an approval.

  2. Synchron obtains an Investigational Device Exemption (IDE) to begin US human trials.

  3. Neuralink performs its first human implant of the N1 device in the PRIME study, placed in the motor cortex by the R1 surgical robot.

    Neuralink — PRIME Study Progress Update (opens in a new tab)
  4. Colorado signs HB 24-1058, the first US state law adding neural data to the protected biological data of its privacy act; opt-in consent is required.

    Effective: 2024-08-07

    ArentFox Schiff — Protections for Neural Data (opens in a new tab)
  5. The New England Journal of Medicine publishes the UC Davis speech neuroprosthesis: attempted speech decoded for a 45-year-old man with ALS from 256 cortical microelectrodes.

    New England Journal of Medicine (2024) (opens in a new tab)
  6. Neuralink receives an FDA Breakthrough Device Designation for "Blindsight," a visual-cortex stimulation application.

    MassDevice — Neuralink FDA breakthrough nod (opens in a new tab)
  7. California signs SB 1223, extending its consumer privacy act to neural data with opt-in consent.

    Effective: 2025-01-01

    Arnold & Porter — Neural Data Privacy Regulation (opens in a new tab)
  8. Synchron reports COMMAND results: 6 patients with severe bilateral upper-limb paralysis, all meeting the primary safety endpoint over 12 months, with the device deployed on target in 100% of cases.

    Synchron / BusinessWire (2024-09-30) (opens in a new tab)
  9. Nature publishes an instantaneous voice-synthesis neuroprosthesis: audible speech generated from raw neural activity in an ALS participant, with closed-loop feedback.

    Nature (2025) — PubMed 40506548 (opens in a new tab)
  10. Precision Neuroscience announces FDA 510(k) clearance for the Layer 7 cortical interface — a 1,024-electrode thin-film surface array cleared to record and stimulate for up to 30 days.

    Precision Neuroscience / GlobeNewswire (opens in a new tab)
  11. Neuralink receives an FDA Breakthrough Device Designation for a speech/communication application.

    MassDevice — Neuralink FDA breakthrough nod (opens in a new tab)
  12. Paradromics places its Connexus device in a human temporarily during an epilepsy surgery, confirming it can be inserted and record signals.

    Michigan Medicine / Paradromics / BusinessWire (opens in a new tab)
  13. By late 2025 Neuralink reports more than a dozen participants and international expansion, with UK (GB-PRIME, NCT07127172) and UAE (UAE-PRIME, NCT06992596) arms on public trial registries — company statements, not peer-reviewed publications.

    Neuralink — PRIME Study Progress Update (opens in a new tab)
  14. Synchron announces a $200M Series D to fund the pivotal trial it says it is preparing.

  15. CorTec's Brain Interchange receives an FDA Breakthrough Device Designation for stroke motor rehabilitation, reported as the first BCI worldwide designated for that indication.

    CorTec / GlobeNewswire (2026-04-08) (opens in a new tab)
  16. The University of Michigan completes the first Connexus implantation for Paradromics' FDA-approved Connect-One early feasibility study.

    Michigan Medicine / Paradromics / BusinessWire (opens in a new tab)

Analysis

Invasiveness is the field's central trade-off, and no design has won

Penetrating microelectrode arrays push needles into tissue and capture the sharpest single-neuron signals; surface arrays sit on top of the cortex and trade resolution for lower surgical risk; Synchron's Stentrode reaches the motor cortex through a vein and avoids opening the skull entirely. More signal generally means more capability and more risk. Three different bets are being placed simultaneously, and none of them has been settled by evidence.

The clearest results are peer-reviewed, and they are about speech

The UC Davis work in the New England Journal of Medicine (2024) and the Nature follow-on (2025) cleared outside review and reported hard numbers — decoding attempted speech, then synthesizing an audible voice from raw neural activity with closed-loop feedback. That is a stronger claim than any press release in this field, which is exactly why the peer-reviewed literature is the right place to look for what has been measured.

Company claims and independent validation carry different weight

Neuralink's cursor performance and typing figures are the company describing its own study; the article's research notes that no peer-reviewed clinical publication of those performance numbers exists as of writing, and that the participant total is given only as a direction rather than a precise figure. Those claims may hold up. Until they are published, they sit on a different rung of the evidence ladder than the NEJM and Nature results.

A regulatory milestone is not a marketing approval

Breakthrough Device Designation speeds up interaction with the FDA for serious conditions; Synchron got one in 2020, Neuralink in September 2024 and May 2025, CorTec in April 2026. Precision's April 2025 clearance covers an electrode array for up to 30 days, not a complete implanted therapeutic system. As of 2026 no fully implanted, general-purpose BCI has FDA marketing approval — every implant discussed is investigational or limited-duration.

Electrode count is a specification, not a score

1,024 surface electrodes, 421 penetrating microelectrodes and 256 microelectrodes across four arrays are not three positions in one race. The devices sit in different places relative to the cortex, record different things, and are being tested for different durations and indications. Counting electrodes across producers is useful for scale, not for ranking.

Small n is the design, not a flaw — but it bounds every conclusion

Six patients in COMMAND and one participant in each speech study are exactly what early feasibility looks like: a first careful look whose primary job is safety. That is how responsible device development starts. It also means nothing here tells us how these systems perform across thousands of diverse patients, how durable the benefit is, or how often the devices fail.

Neural-data law arrived before the consumer market did

Colorado added neural data to its privacy act in 2024 and California extended its consumer privacy act effective January 2025, both requiring opt-in consent, with more states introducing similar bills. Supporters call it pre-emption of a real harm; skeptics note the definitions are new and untested and that medical-research data is often handled under separate rules. Either way, the rulemaking is running ahead of the products.

Reversibility and long-term safety are the unwritten chapter

An implant is not a phone you upgrade — removing or replacing one is surgery, and tissue can react to a foreign body over years. Multi-year data simply does not exist yet for most of these devices, which is why early studies emphasize safety endpoints and why less invasive designs trade capability for lower explantation risk.

Comparison

What each result actually establishes. Evidence tier follows the article's research notes; "scale" is study size or clearance scope, not performance.
ProgramEvidence tierScaleWhat is not yet established
Neuralink PRIME (cursor, typing)Company statementMore than a dozen participants, expanding; no precise total publishedNo peer-reviewed clinical publication of the performance figures
Synchron COMMANDProspective safety result, company-reported6 patients, 12-month evaluationEfficacy described qualitatively, not proven in a controlled trial
UC Davis speech decoding (NEJM 2024)Peer-reviewed1 participantA population-level result; this is proof of concept
Voice synthesis (Nature 2025)Peer-reviewed1 participantDurability and performance beyond a single case
Precision Layer 7FDA 510(k) clearanceRecord and stimulate for up to 30 daysApproval of a complete, permanently implanted therapeutic system
Paradromics ConnexusInstitution-reported feasibilityFirst implantation for an FDA-approved EFS (June 2026)Any efficacy outcome
CorTec Brain InterchangeBreakthrough Device DesignationStroke motor rehabilitation indicationDesignation is a fast lane, not an approval
Three routes to the same cortex, with different surgical bargains.
ApproachExampleSignalSurgical trade-off
Penetrating microelectrode arrayUtah Array (BrainGate); Neuralink N1; Paradromics Connexus (421 electrodes)Sharpest — individual neuronsNeedles enter brain tissue; removal or replacement is surgery
Surface arrayPrecision Neuroscience Layer 7 (1,024 electrodes)Some resolution given upLies on the cortex without piercing it; lower surgical risk
EndovascularSynchron StentrodeCoarsest of the threeThreaded up the jugular vein; no open-skull surgery
The US pathway, and where each program currently sits.
StepWhat it meansWho is there
Breakthrough Device DesignationFaster interaction with the FDA for serious conditions — not approvalSynchron (2020), Neuralink (Sept 2024, May 2025), CorTec (Apr 2026)
IDE + early feasibility studyPermission to implant in a small number of people, safety firstSynchron COMMAND (6 patients); Paradromics Connect-One
510(k) clearanceClearance of a component or limited-duration usePrecision Layer 7 (up to 30 days)
Pivotal trialThe larger confirmatory trial before a marketing applicationSynchron has said it is preparing one
Premarket approval (PMA)Permission to sell the deviceAs of 2026, no fully implanted general-purpose BCI has it
The first two US neural-data statutes. Both generally require opt-in consent to collect and process neural data, and several more states have introduced similar bills.
StateLawSignedEffectiveWhat it does
ColoradoHB 24-10582024-04-172024-08-07First US state law adding neural data to protected biological data under its privacy act
CaliforniaSB 12232024-09-272025-01-01Extends the consumer privacy act's protections to neural data

Process

  1. Breakthrough Device Designation

    A faster lane for serious conditions. Not an approval — Synchron 2020, Neuralink 2024 and 2025, CorTec 2026.

  2. Investigational Device Exemption (IDE)

    The permission required to implant an unapproved device in people at all. Synchron obtained one in 2021.

  3. Early feasibility study

    A small, safety-first trial. COMMAND enrolled 6 patients; Paradromics' Connect-One began with a first implantation in June 2026.

  4. Pivotal trial

    The larger confirmatory study. Synchron has said it is preparing one; none of these implants has completed this step.

  5. Premarket approval (PMA)

    The point at which a device can be sold. As of 2026 no fully implanted, general-purpose BCI has reached it.

Sources

  1. New England Journal of Medicine — An Accurate and Rapidly Calibrating Speech Neuroprosthesis (2024).View source (opens in a new tab)
  2. UC Davis Health — New brain-computer interface allows man with ALS to 'speak' again (2024-08-14).View source (opens in a new tab)
  3. Nature — An instantaneous voice-synthesis neuroprosthesis (2025).View source (opens in a new tab)
  4. Synchron — Positive Results from the U.S. COMMAND Study of an Endovascular Brain-Computer Interface (2024-09-30).View source (opens in a new tab)
  5. Neuralink — PRIME Study Progress Update (2025).View source (opens in a new tab)
  6. MassDevice — Neuralink wins FDA breakthrough nod for speech-impairment BCI (2025).View source (opens in a new tab)
  7. Precision Neuroscience — FDA Clearance for High-Resolution Cortical Electrode Array (Layer 7) (2025-04-17).View source (opens in a new tab)
  8. Michigan Medicine / Paradromics — First Connexus BCI Implantation for the FDA-Approved Connect-One Study (2026-06-17).View source (opens in a new tab)
  9. ClinicalTrials.gov — GB-PRIME (NCT07127172) and UAE-PRIME (NCT06992596) (2025).View source (opens in a new tab)
  10. ArentFox Schiff — California and Colorado Establish Protections for Neural Data (2024).View source (opens in a new tab)
  11. Arnold & Porter — Neural Data Privacy Regulation: What Laws Exist and What Is Anticipated? (2025-07).View source (opens in a new tab)
  12. CorTec — FDA Breakthrough Device Designation for Brain Interchange in Stroke Rehabilitation (2026-04-08).View source (opens in a new tab)

Tags

  • #brain-computer-interface
  • #neuralink
  • #synchron
  • #neural-implants
  • #medical-devices