Brain–computer interface: revision 2 to 6

Revision 2

Revision 6

type: "technology"

status: "emerging"

horizon: "late 2020s"

trl: 6

trl: 7

categories: ["cybernetics", "foundations"]

tags: ["bci", "neurotechnology", "paralysis", "neural decoding", "implants", "assistive technology"]

summary: "A system that measures nervous-system activity, decodes an intention or state from it, and converts that decoded signal into a command for an external device."

summary: "A system that measures nervous-system activity, decodes an intention or state from it, and converts that decoded signal into a command for an external device."

updated: "2026-07-27"

updated: "2026-10-30"

humanEvidence: "Implanted arrays have given research participants with paralysis cursor control, robotic-arm use and conversational-rate speech decoding; no implanted BCI has been shown to improve any capability in a healthy person."

humanEvidence: "Implanted arrays have given research participants with paralysis cursor control, robotic-arm use and large-vocabulary speech decoding at up to about 60-80 words per minute; no implanted BCI has been shown to improve any capability in a healthy person."

access: "No implanted BCI is approved by any regulator; the only route is enrolment in an early feasibility trial. Consumer EEG headsets are sold freely as wellness devices."

access: "China approved one epidural BCI for hand grasp after cervical spinal cord injury in March 2026; elsewhere implanted BCIs are trial-only. Consumer EEG headsets are sold freely as wellness devices."

reversibility: "difficult"

issues: ["The speech-decoding section states vocabulary and error-rate figures without citations."]

issues: ["Trial statuses and participant counts in Current state are as of October 2026 and need rechecking as studies report.", "Chinese BCI programmes other than the approved Neuracle device are covered only in passing."]

---

```infobox

{ "label": "Signal sources", "value": "Spikes, LFP, ECoG, EEG, fNIRS" },

{ "label": "First human implant trial", "value": "2004 (BrainGate)" },

{ "label": "Primary clinical use", "value": "Severe paralysis, anarthria" },

{ "label": "Regulatory status", "value": "Investigational; no approved implanted BCI" },

{ "label": "Market approval", "value": "2026, China (Neuracle NEO, hand grasp)" },

{ "label": "Readiness", "value": "TRL 6" }

{ "label": "Regulatory status", "value": "Investigational outside China" },

{ "label": "Readiness", "value": "TRL 7" }

]

}

```

**Brain–computer interface** (BCI) is the general term for a system that measures activity in the

**Brain-computer interface** (BCI) is the general term for a system that measures activity in the

nervous system, extracts a variable of interest from that activity, and turns it into a command for

a computer, a prosthesis, or the person's own muscles. The defining feature, in the standard

formulation, is that the command bypasses the ordinary output path of peripheral nerve and

muscle.[^wolpaw2002] A BCI is therefore useful mainly to people

muscle.[^wolpaw2002] A BCI is therefore useful mainly to people whose ordinary output path is

whose ordinary output path is broken, and as of 2026 every serious clinical result comes from that

broken, and as of October 2026 every serious clinical result comes from that population: participants with

population: participants with tetraplegia, brainstem stroke, or amyotrophic lateral sclerosis.

tetraplegia, brainstem stroke, or amyotrophic lateral sclerosis.

The field has produced genuinely striking demonstrations — a person with no usable movement

The field has produced striking demonstrations in a very small number of research participants: a

controlling a robotic arm, another producing fluent synthesized speech from motor-cortex activity

person with no usable movement controlling a robotic arm, another producing fluent synthesized

alone — in a very small number of research participants, using hardware that is tethered, requires

speech from motor-cortex activity alone. The hardware behind them is tethered, requires frequent

frequent recalibration, and has not been approved by any regulator for permanent use. The gap

recalibration, and has not been approved by any regulator for permanent use. The implanted BCI for

between those demonstrations and a product is largely a gap in reliability, longevity, and

movement that reached the market in China in March 2026 is far simpler: eight electrodes resting on

manufacturing, not in principle.

the dura that let a person with a cervical spinal cord injury close a pneumatic glove.[^stdaily2026][^cyd2026] The

gap between the laboratory demonstrations and a product lies largely in reliability, longevity, and

manufacturing rather than in principle.

```figure

{"key": "braingate-array", "caption": "The implanted array against a human hand. The entire recording surface is a few millimetres square."}

{"key": "braingate-array", "caption": "The implanted array against a human hand. The entire recording surface is a few millimetres square."}

## The signal chain

Every BCI is the same four stages: acquisition, feature extraction, decoding, and output, usually

Every BCI has the same four stages: acquisition, feature extraction, decoding, and output, usually

with a feedback loop closing back to the user.

*Acquisition* determines everything downstream. Signals differ by how close the sensor sits to the

neurons producing them. Penetrating microelectrodes recording action potentials from individual

cells carry the most information per channel; scalp electroencephalography, separated from cortex

cells carry the most information per channel; scalp electroencephalography, separated from cortex by

by dura, skull, and scalp, carries the least, because the skull acts as a spatial low-pass filter

dura, skull, and scalp, carries the least, because the skull acts as a spatial low-pass filter that

that smears contributions from many square centimetres of cortex into one measurement.

smears contributions from many square centimetres of cortex into one measurement.

*Feature extraction* reduces raw voltage into something a decoder can use: firing rates binned over

tens of milliseconds, power in a frequency band, the amplitude of an evoked potential. In cortical

recordings the workhorse features are threshold crossings and high-gamma power (roughly 70–200 Hz),

recordings the workhorse features are threshold crossings and high-gamma power (roughly 70-200 Hz),

the latter being a reasonable proxy for local population spiking that can be measured without

penetrating the cortex.

*Decoding* maps features to intent. Early systems used linear filters and the population-vector

approach inherited from primate motor physiology; current systems use Kalman filters, recurrent

neural networks, and increasingly transformer-style sequence models paired with [language models](https://aiwiki.ai/wiki/large_language_model).

neural networks, and increasingly transformer-style sequence models paired with [language

[[neural-decoding]] is where most of the recent performance gains have come from — the electrodes

models](https://aiwiki.ai/wiki/large_language_model). Most of the recent performance gains have come

in the best 2025 speech systems are not fundamentally better than those used in 2012.

from [[neural-decoding]]; the electrodes in the best 2025 speech systems are not fundamentally

better than those used in 2012.

*Output* is a cursor, a robotic arm, a synthesized voice, a wheelchair, or stimulation delivered to

the user's own spinal cord or to a powered [[exoskeleton]]. Bidirectional systems also write

a character grid and detects the evoked response that follows an attended flash; steady-state

visually evoked potential systems tag each option with a distinct flicker frequency and read off

which frequency dominates visual cortex; sensorimotor-rhythm systems ask the user to imagine

movement of a limb and detect the resulting suppression of mu and beta rhythms. All three work.

movement of a limb and detect the resulting suppression of mu and beta rhythms. All three work, but

None is fast: information transfer rates are conventionally reported in the range of tens of bits

none is fast: information transfer rates are conventionally reported in the range of tens of bits

per minute, on the order of a few characters, against several hundred bits per minute for competent

typing.

Functional near-infrared spectroscopy and functional MRI measure haemodynamics rather than

electrical activity, and are therefore limited by the seconds-long delay of the blood-oxygen

response. fMRI remains valuable for decoding research precisely because it covers the whole brain,

response. fMRI remains valuable for decoding research because it covers the whole brain, but a

but a scanner is not a wearable device. Consumer EEG headsets sit at the weakest end

scanner is not a wearable device. Time-domain fNIRS, which times returning photons to gain depth

of the spectrum, typically with a handful of dry electrodes and heavy artefact contamination from

sensitivity, has been built into a whole-head wearable helmet by [[kernel]], whose first-generation

eye movement and jaw muscle.

Kernel Flow weighed about 2 kg and was benchmarked mainly on optical phantoms and a two-person

finger-tapping task.[^ban2022] Kernel sells it for research and brain-health measurement rather than

for device control. Consumer EEG headsets sit at the weakest end of the spectrum, typically with a

handful of dry electrodes and heavy artefact contamination from eye movement and jaw muscle.

```compare

{

"rows": [

    { "label": "Surgery", "values": ["Craniotomy, cortical penetration", "Craniotomy or slot", "Catheter via jugular vein", "None"] },

    { "label": "Surgery", "values": ["Craniotomy, cortical penetration", "Craniotomy or slot", "Catheter via jugular vein", "None"] },

    { "label": "Resolution", "values": ["Single neurons", "Millimetre populations", "Coarse populations", "Centimetre populations"] },

    { "label": "Resolution", "values": ["Single neurons", "Millimetre populations", "Coarse populations", "Centimetre populations"] },

{ "label": "Typical channels", "values": ["96–1,024", "64–1,024", "16", "8–256"] },

{ "label": "Typical channels", "values": ["96-1,024", "64-1,024", "16", "8-256"] },

    { "label": "Long-term stability", "values": ["Degrades over months to years", "Good over years", "Good over years", "Good, but user-dependent"] },

    { "label": "Long-term stability", "values": ["Degrades over months to years", "Good over years", "Good over years", "Good, but user-dependent"] },

    { "label": "Demonstrated ceiling", "values": ["Multi-degree-of-freedom limb control, fluent speech", "Fluent speech, walking via spinal stimulation", "Discrete clicks", "Spellers, simple switches"] }

    { "label": "Demonstrated ceiling", "values": ["Multi-degree-of-freedom limb control, fluent speech", "Fluent speech, walking via spinal stimulation", "Discrete clicks", "Spellers, simple switches"] }

]

[

  { "year": "1924", "title": "Human EEG recorded", "text": "Hans Berger records electrical rhythms from the human scalp, establishing that cortical activity can be measured non-invasively." },

  { "year": "1924", "title": "Human EEG recorded", "text": "Hans Berger records electrical rhythms from the human scalp, establishing that cortical activity can be measured non-invasively." },

  { "year": "1973", "title": "The term is coined", "text": "Jacques Vidal at UCLA publishes 'Toward direct brain-computer communication', proposing that computers could interpret EEG in real time as control signals." },

  { "year": "1973", "title": "The term is coined", "text": "Jacques Vidal at UCLA publishes 'Toward direct brain-computer communication', proposing that computers could interpret EEG in real time as control signals." },

  { "year": "1999–2002", "title": "Motor decoding in animals", "text": "Groups led by John Chapin, Miguel Nicolelis, and Andrew Schwartz show that rats and monkeys can control levers and robotic arms using decoded cortical ensemble activity." },

  { "year": "1999-2002", "title": "Motor decoding in animals", "text": "Groups led by John Chapin, Miguel Nicolelis, and Andrew Schwartz show that rats and monkeys can control levers and robotic arms using decoded cortical ensemble activity." },

  { "year": "2004–2006", "title": "First long-term human implant", "text": "Matthew Nagle, paralysed by a spinal cord injury, receives a Utah array in the BrainGate pilot and controls a cursor, a television, and a prosthetic hand." },

  { "year": "2004-2006", "title": "First long-term human implant", "text": "Matthew Nagle, paralysed by a spinal cord injury, receives a Utah array in the BrainGate pilot and controls a cursor, a television, and a prosthetic hand." },

  { "year": "2012–2013", "title": "Robotic arm control", "text": "BrainGate and University of Pittsburgh teams report multi-degree-of-freedom robotic arm control by participants with tetraplegia, including self-feeding." },

  { "year": "2012-2013", "title": "Robotic arm control", "text": "BrainGate and University of Pittsburgh teams report multi-degree-of-freedom robotic arm control by participants with tetraplegia, including self-feeding." },

  { "year": "2016", "title": "Fully implanted communication BCI", "text": "A Utrecht group implants a wireless electrocorticography system in a person with late-stage ALS, who uses it at home for years to select letters." },

  { "year": "2016", "title": "Fully implanted communication BCI", "text": "A Utrecht group implants a wireless electrocorticography system in a person with late-stage ALS, who uses it at home for years to select letters." },

  { "year": "2021–2024", "title": "Speech decoding becomes fluent", "text": "UCSF, Stanford, and UC Davis teams decode attempted speech at conversational rates from motor cortex, with large vocabularies and single-digit word error rates in the best case." },

  { "year": "2021-2024", "title": "Speech decoding scales up", "text": "UCSF, Stanford, and UC Davis teams decode attempted speech from sensorimotor cortex at up to 62-78 words per minute over large vocabularies, with single-digit word error rates in the best case." },

  { "year": "2023", "title": "Brain–spine interface", "text": "A Lausanne team restores volitional walking in a man with spinal cord injury by linking a cortical implant to an epidural spinal stimulator." },

  { "year": "2023", "title": "Brain-spine interface", "text": "A Lausanne team restores volitional walking in a man with spinal cord injury by linking a cortical implant to an epidural spinal stimulator." },

  { "year": "2024–2026", "title": "Commercial implants enter trials", "text": "Neuralink, Synchron, Precision Neuroscience, Paradromics, and several Chinese groups run early feasibility studies; none has regulatory approval for routine use." }

  { "year": "2024-2026", "title": "Commercial implants enter trials", "text": "Neuralink, Synchron, Precision Neuroscience, Paradromics, and several Chinese groups run early feasibility studies. No implanted BCI is approved in the US; high-density cortical arrays such as Precision's Layer 7-T are cleared only as electrodes for temporary recording and stimulation of under 30 days." },

  { "year": "2026", "title": "Market approval in China", "text": "China's National Medical Products Administration approves Neuracle's NEO, an epidural BCI for hand grasp after cervical spinal cord injury, in March; the first implant outside a trial follows in July." }

]

```

**Cursor and device control.** People with tetraplegia can move a computer pointer, click, type on a

virtual keyboard, and drive a robotic arm through several degrees of freedom.[^hochberg2006] The

2012 and 2013 robotic-arm reports, in which participants reached, grasped, and drank

2012 and 2013 robotic-arm reports, in which participants reached, grasped, and drank unassisted,

unassisted, remain the reference demonstrations.[^hochberg2012][^collinger2013] Performance in the

remain the reference demonstrations.[^hochberg2012][^collinger2013] Performance in the best

best participants approaches a slow but usable pointing device. The [[utah-array]] remains the most

participants approaches a slow but usable pointing device. The [[utah-array]] remains the most

common sensor for this work, and the multi-site BrainGate consortium has produced most of the

long-run data. A fully implanted surface-electrode system used at home by a person with late-stage

ALS demonstrated that the approach can survive outside a laboratory.[^vansteensel2016]

**Speech.** Between 2021 and 2025 decoding of attempted speech moved from a 50-word vocabulary at

**Speech.** In 2021 a subdural array decoded attempted speech from a 50-word vocabulary at a median

around fifteen words per minute to conversational-rate output over vocabularies of more than a

15.2 words per minute, with a median word error rate of 25.6%.[^moses2021] Two 2023 reports moved to

hundred thousand words, with the best reported word error rates in the single digits for one

large vocabularies: one, using intracortical arrays, reached 62 words per minute at a 23.8% word

participant. Some systems now synthesize a voice reconstructed from pre-illness recordings, or

error rate over 125,000 words,[^willett2023] and the other, using a surface array, 78 words per

animate an avatar. See [[speech-neuroprosthesis]].

minute at a 25% error rate.[^metzger2023] In 2024 a man with ALS and four intracortical arrays

sustained 97.5% word accuracy over that vocabulary and conversed at about 32 words per

minute,[^card2024] and a 2026 follow-up reported him using it at home over nearly two years for more

than 3,800 hours, with no researchers present.[^card2026] Some systems now speak in a voice modelled on

the person's pre-illness recordings, or animate an avatar. All of these results come from one

participant each. See [[speech-neuroprosthesis]].

**Movement restoration.** Rather than driving an external robot, a cortical implant can drive

stimulation of the user's own spinal cord or muscles, re-establishing a "digital bridge" around a

([[neuralink]], Paradromics, Blackrock Neurotech), thin-film surface arrays ([[ecog-interfaces]],

Precision Neuroscience), and endovascular electrodes threaded through a vein ([[stentrode]],

developed by [[synchron]]). Each trades bandwidth against surgical risk. Several Chinese academic

and state-backed groups have also reported implants in participants since 2024, though public

and state-backed groups have also reported implants in participants since 2024, though public detail

detail is limited and independent verification scarce.

is limited and independent verification scarce.

What has not been demonstrated is equally worth stating. No BCI has restored a rich sense of touch;

No BCI has restored a rich sense of touch; the best feedback so far is a small number of

the best feedback so far is a small number of discriminable pressure percepts delivered by

discriminable pressure percepts delivered by intracortical or peripheral stimulation, well short of

intracortical or peripheral stimulation, well short of what a hand needs. No system has restored

what a hand needs. Vision by cortical stimulation is earlier still: in October 2026 Elon Musk

memory in a clinically meaningful way, despite the encoding-model work described under

described Neuralink's Blindsight implant as awaiting regulatory approval (see

[[memory-prosthesis]]. And no implanted BCI has been shown to improve any capability in a healthy

[[cortical-visual-prosthesis]]).[^musk2026b] No system has restored memory in a clinically

person, which is the premise on which most enhancement-oriented interest rests.

meaningful way, despite the encoding-model work described under [[memory-prosthesis]]. Nor has any

implanted BCI been shown to improve any capability in a healthy person, which is the premise on

which most enhancement-oriented interest rests.

> [!caution] Sample sizes are tiny

> The headline results in this field typically come from one to four participants, often the same

> been tested at the scale that would establish reliability, and single-participant results have

> repeatedly failed to generalize.

### Regulatory status as of October 2026

On 13 March 2026 China's National Medical Products Administration registered Neuracle's NEO as a

Class III device for adults with cervical spinal cord injury who cannot grasp. The coin-sized wireless implant sits outside the dura, and its decoded signals drive a

pneumatic glove.[^stdaily2026] Its developers report that all 32 patients in a

confirmatory trial achieved brain-controlled grasp; Hong Bo, whose Tsinghua University team

designed it, cautioned that a first approved product did not put China at the

front of the field.[^cyd2026] The first implant outside a trial followed in July 2026 in Shanghai, and the city has added the device to its Huhuibao supplementary insurance

scheme.[^gmw2026] The approval followed a 2025 industrial policy from seven agencies that aims for two to three

globally influential BCI companies by 2030.[^miit2025] Chinese media had already called a Sceneray addiction

treatment registered in December 2025 an invasive BCI, so "first" claims for NEO depend on

definitions.[^cls2025]

By October 2026 the US Food and Drug Administration had approved no implanted BCI. Precision

Neuroscience's 1,024-electrode Layer 7-T array was cleared in March 2025 as a cortical electrode for

recording and stimulation, not as a BCI, and only for temporary use of under 30 days.[^fda2025] Paradromics gained an investigational device

exemption for its Connect-One speech study in November 2025 and implanted its first participant in

June 2026; it says the participant now communicates by synthesized speech, text messages and phone

calls.[^paradromics2026] Synchron reported in 2024 that its six-person COMMAND study met its primary

safety endpoint,[^evtoday2024] and raised $200 million in November 2025 partly to prepare a

pivotal trial.[^medtechdive2025] No such trial was registered on ClinicalTrials.gov by October 2026;

its newest US study, INTENT, is another early feasibility study, enrolling since May

2026.[^intent2026] Neuralink said in January 2026 that 21 people were enrolled in its trials

worldwide.[^reuters2026]

## The bandwidth problem

Popular framing treats BCI progress as an electrode-count race, on the assumption that information

transfer scales with channels. It does not scale cleanly. Motor cortical activity during reaching

transfer scales with channels. It does not scale cleanly. Motor cortical activity during reaching is

is well described by a low-dimensional dynamical structure — a neural manifold — in which a few

well described by a low-dimensional dynamical structure (a neural manifold) in which a few dozen

dozen latent dimensions capture most of the variance.[^gallego2017] Adding electrodes samples that same structure

latent dimensions capture most of the variance.[^gallego2017] Adding electrodes samples that same

more densely rather than revealing new independent signals, so returns diminish. This is why a

structure more densely rather than revealing new independent signals, so returns diminish. This is

96-channel array from the 1990s can support performance comparable to systems with ten times the

why a 96-channel array from the 1990s can support performance comparable to systems with ten times

channels, and why decoder architecture has mattered more than sensor count.

the channels, and why decoder architecture has mattered more than sensor count.

The fastest peer-reviewed large-vocabulary BCI communication is attempted speech, at 62-78 words

per minute,[^willett2023][^metzger2023] against about 160 for ordinary conversation;[^willett2023] attempted handwriting has reached about 90

characters per minute.[^willett2021] Jieyu Zheng and Markus Meister put the information throughput

of human behaviour at about 10 bits per second, with tasks from typing to speech falling between

roughly 5 and 50, and place the limit in cognition rather than in the fingers or mouth; they predict

that Musk himself, given a Neuralink implant, would communicate at about that rate.[^zheng2025] Musk

forecast otherwise on 30 October 2026: the output bandwidth of Neuralink's Telepathy product "will

vastly exceed the fastest human speech or typing in a few years".[^musk2026] No published BCI result

has yet matched conversational speech, let alone the fastest typists.

Where more channels do help is coverage: sampling several cortical areas, or sampling a functional

map finely enough to separate articulators in speech cortex. The genuinely hard constraints are

map finely enough to separate articulators in speech cortex. The hard constraints lie elsewhere: the

elsewhere — the number of independent control dimensions a person can learn to modulate, the

number of independent control dimensions a person can learn to modulate, the non-stationarity of the

non-stationarity of the recorded population from day to day, and the near-total absence of a

recorded population from day to day, and the near-total absence of a high-resolution write channel.

high-resolution write channel. Reading intention out is a solved-enough problem for many

Reading intention out is a solved-enough problem for many applications; writing rich sensory

applications; writing rich sensory information in is not, which is why touch feedback in

information in is not, which is why touch feedback in [[neuroprosthetics]] remains crude compared

[[neuroprosthetics]] remains crude compared with motor decoding.

with motor decoding. One laboratory response is to make the stimulating hardware itself spike like a

neuron, at biological voltages and timescales; such

[[artificial-neurons|bio-interfacing artificial neurons]] have so far driven living neurons only in

acute experiments, such as slices of mouse cerebellum.[^hadke2026]

## Limitations

Implanted arrays lose channels. The brain mounts a foreign-body response — microglial activation,

Implanted arrays lose channels. The brain mounts a foreign-body response (microglial activation,

astrocytic encapsulation — that displaces neurons from the recording tip, while insulation and

astrocytic encapsulation) that displaces neurons from the recording tip, while insulation and metal

metal traces degrade in a warm saline environment. Signal yield typically declines over months to

traces degrade in a warm saline environment. Signal yield typically declines over months to years,

years, though some implants have supported usable control beyond a thousand days and a few for

though some implants have supported usable control beyond a thousand days and a few for several

several years; the failure statistics are set out under [[utah-array]].

years; the failure statistics are set out under [[utah-array]].

Percutaneous connectors, still standard in academic systems, carry infection risk and tether the

user to a lab. Fully implanted wireless systems solve this but constrain power and data rate; the

most aggressive proposal for removing the tether, untethered ultrasonic motes, is covered under

[[neural-dust]] and has no human results.

[[neural-dust]] and has no human results. Decoders drift, requiring recalibration that ranges from

Decoders drift, requiring recalibration that ranges from minutes daily to occasional, depending on

minutes daily to occasional, depending on the system. Almost all reported performance comes from

the system. Almost all reported performance comes from controlled settings with a technician

controlled settings with a technician present; the home-use reports described above are the

present.

exceptions.

There is also a structural risk specific to commercial neurotechnology: a company can fail while

There is also a structural risk specific to commercial neurotechnology: a company can fail while its

its devices remain in people's heads. The discontinuation of the Argus II retinal system left users

devices remain in people's heads. The discontinuation of the Argus II retinal system left users with

with unsupported implants, a precedent that shadows every BCI trial. See [[retinal-implant]].

unsupported implants, a precedent that shadows every BCI trial. See [[retinal-implant]].

## Ethics and governance

Implanted BCIs raise questions that older neurotechnology such as [[deep-brain-stimulation]] and

Implanted BCIs raise questions that older neurotechnology such as [[deep-brain-stimulation]] and the

the [[cochlear-implant]] introduced in milder form: who owns the neural data, what happens when a

[[cochlear-implant]] introduced in milder form: who owns the neural data, what happens when a

decoder misreads intent, and whether a device that shapes behaviour compromises agency. The

decoder misreads intent, and whether a device that shapes behaviour compromises agency. The specific

specific novelty is that recorded activity can support inferences the user did not intend to

novelty is that recorded activity can support inferences the user did not intend to disclose, an

disclose, an issue taken up under [[mental-privacy]]. Legislatures have begun responding —

issue taken up under [[mental-privacy]]. Legislatures have begun responding under the banner of

constitutional and state-level neural-data provisions exist in several jurisdictions — under the

[[neurorights]], and constitutional and state-level neural-data provisions exist in several

banner of [[neurorights]], though enforcement mechanisms remain thin.

jurisdictions, though enforcement mechanisms remain thin.

Consumer neurotechnology is a separate regulatory category and a much weaker technical one.

Consumer neurotechnology is a separate regulatory category and a much weaker technical one. Wearable

Wearable EEG headsets and [[non-invasive-neuromodulation]] devices are sold for attention training,

EEG headsets and [[non-invasive-neuromodulation]] devices are sold for attention training, sleep,

sleep, and gaming with evidence that ranges from thin to absent, and the "brain-reading" claims

and gaming with evidence that ranges from thin to absent, and the "brain-reading" claims attached to

attached to them substantially outrun what scalp electrodes can resolve. Regulators have generally

them substantially outrun what scalp electrodes can resolve. Regulators have generally treated these

treated these as wellness products rather than medical devices, the same line that governs

as wellness products rather than medical devices, the same line that governs

[[wearable-health-sensors]], which is why the emerging neural-data statutes target data handling

rather than efficacy claims.

> [!debate] How invasive is worth it

> Synchron's position is that a device delivered by catheter, with sixteen electrodes, will reach

> far more patients than one requiring a craniotomy, and that discrete reliable control is enough

> for most assistive tasks. The counter-position, taken by the penetrating-array developers, is

> for most assistive tasks. The counter-position, taken by the penetrating-array developers, is that

> that low-bandwidth control caps the addressable applications at switch-like interaction and rules

> low-bandwidth control caps the addressable applications at switch-like interaction and rules out

> out speech and dexterous limb control. Both claims are testable and neither has been settled.

> speech and dexterous limb control. Both claims are testable and neither has been settled.

## Outlook

The near-term agenda is legible: more participants, wireless and fully implanted systems, decoders

The near-term agenda is clear: more participants, wireless and fully implanted systems, decoders

intended to survive a week without recalibration, and eventually a pivotal trial in a well-defined

indication such as loss of speech in ALS. If that sequence holds, it could produce an approved

indication such as loss of speech in ALS. If that sequence holds, it could produce a US-approved

assistive device for a small clinical population, though no timeline for it is well supported and

assistive device for a small clinical population, as China has already approved one for hand

none of the current studies is large enough to predict one.

grasp, though no US timeline is well supported and none of the current studies is large enough to

predict one.

The longer-term claims — that BCIs will serve as a general-purpose channel for

Longer-term claims rest on assumptions the clinical work does not test: that BCIs will serve as a

[[human-ai-merger]], or provide the read-out fidelity that [[whole-brain-emulation]] would require

general-purpose channel for [[human-ai-merger]], or provide the read-out fidelity that

— rest on assumptions the clinical work does not test. A speech decoder recovers a motor plan, not

[[whole-brain-emulation]] would require. A speech decoder recovers a motor plan, not a thought; a

a thought; a 1,024-channel array samples a millionth of a cortical hemisphere. Whether elective

1,024-channel array samples a millionth of a cortical hemisphere. Whether elective implantation in

implantation in healthy people ever clears the risk-benefit bar is a question about surgical

healthy people ever clears the risk-benefit bar is a question about surgical complication rates and

complication rates and device lifetimes, and neither is currently known well enough to answer.

device lifetimes, and neither is currently known well enough to answer.

## See also

[^lorach2023]: `paper` Lorach, H. et al. "Walking naturally after spinal cord injury using a brain–spine interface." *Nature*, 2023.

[^lorach2023]: `paper` Lorach, H. et al. "Walking naturally after spinal cord injury using a brain–spine interface." *Nature*, 2023.

[^wolpaw2002]: `paper` Wolpaw, J. R. et al. "Brain–computer interfaces for communication and control." *Clinical Neurophysiology*, 2002.

[^wolpaw2002]: `paper` Wolpaw, J. R. et al. "Brain–computer interfaces for communication and control." *Clinical Neurophysiology*, 2002.

[^hadke2026]: `paper` Hadke, S. S. et al. "Printed MoS2 memristive nanosheet networks for spiking neurons with multi-order complexity." *Nature Nanotechnology*, 2026. {Stimulation of Purkinje neurons in mouse cerebellar slices; no living animal or implanted device.}

[^ban2022]: `paper` Ban, H. Y. et al. "Kernel Flow: a high channel count scalable time-domain functional near-infrared spectroscopy system." *Journal of Biomedical Optics* 27(7):074710, 2022. https://doi.org/10.1117/1.JBO.27.7.074710

[^moses2021]: `paper` Moses, D. A. et al. "Neuroprosthesis for Decoding Speech in a Paralyzed Person with Anarthria." *New England Journal of Medicine* 385(3):217-227, 2021. https://doi.org/10.1056/NEJMoa2027540 {One participant with anarthria after brainstem stroke, recorded with a subdural array over sensorimotor cortex.}

[^willett2023]: `paper` Willett, F. R. et al. "A high-performance speech neuroprosthesis." *Nature* 620:1031-1036, 2023. https://doi.org/10.1038/s41586-023-06377-x {One participant with ALS; the 160 words per minute conversational figure is quoted from earlier literature, not measured in this study.}

[^metzger2023]: `paper` Metzger, S. L. et al. "A high-performance neuroprosthesis for speech decoding and avatar control." *Nature* 620:1037-1046, 2023. https://doi.org/10.1038/s41586-023-06443-4

[^card2024]: `paper` Card, N. S. et al. "An Accurate and Rapidly Calibrating Speech Neuroprosthesis." *New England Journal of Medicine* 391(7):609-618, 2024. https://doi.org/10.1056/NEJMoa2314132

[^card2026]: `paper` Card, N. S. et al. "Long-term independent use of an intracortical brain-computer interface for speech and cursor control." *Nature Medicine* 32:2504-2510, 2026. https://doi.org/10.1038/s41591-026-04414-6 {Same single participant as the 2024 report; the system still used percutaneous wired connectors.}

[^willett2021]: `paper` Willett, F. R. et al. "High-performance brain-to-text communication via handwriting." *Nature* 593:249-254, 2021. https://doi.org/10.1038/s41586-021-03506-2

[^zheng2025]: `paper` Zheng, J. and Meister, M. "The unbearable slowness of being: Why do we live at 10 bits/s?" *Neuron* 113(2):192-204, 2025. https://doi.org/10.1016/j.neuron.2024.11.008 {A perspective that aggregates published behavioural estimates; it reports no new experiment.}

[^musk2026]: `statement` Musk, E. Post on X, 30 October 2026. https://x.com/elonmusk/status/2116018780111597911

[^musk2026b]: `statement` Musk, E. Post on X, 30 October 2026. https://x.com/elonmusk/status/2116019939056513218

[^stdaily2026]: `news` "全球首个植入式脑机接口三类医疗器械获批上市" (First implantable BCI Class III medical device approved for market). *Science and Technology Daily*, 13 March 2026. https://www.stdaily.com/web/gdxw/2026-03/13/content_485324.html

[^cyd2026]: `news` Yang, J., Tao, T. and Yuan, J. "一款侵入式脑机接口医疗器械获批上市" (An invasive BCI medical device approved for market). *China Youth Daily*, 28 March 2026, reprinted by Tsinghua University News. https://www.tsinghua.edu.cn/info/1182/125032.htm {Trial outcomes are as described by the developers; no peer-reviewed report of the confirmatory trial is cited.}

[^cls2025]: `news` Cailian Press. "我国又一侵入式脑机接口产品获批" (Another Chinese invasive BCI product approved). 17 December 2025. https://www.cls.cn/detail/2231624 {A financial news brief; it reports the NMPA Class III registration for Sceneray's addiction device without technical detail.}

[^gmw2026]: `news` Jiang, H. "上海开出脑机接口首张处方并成功手术" (Shanghai issues first BCI prescription and completes surgery). *People's Daily* app Shanghai channel, reprinted by Guangming Online, 15 July 2026. https://m.gmw.cn/2026-07/15/content_1304532828.htm

[^miit2025]: `regulator` Ministry of Industry and Information Technology and six other bodies. "Implementation Opinions on Promoting the Innovative Development of the Brain-Computer Interface Industry." MIIT Document No. 164 (2025); English translation by the Center for Security and Emerging Technology. https://cset.georgetown.edu/wp-content/uploads/t0659_BCI_opinions_EN.pdf

[^fda2025]: `regulator` US Food and Drug Administration. 510(k) K242618, Layer 7-T cortical electrodes, Precision Neuroscience, decision 30 March 2025. https://www.accessdata.fda.gov/cdrh_docs/pdf24/K242618.pdf

[^paradromics2026]: `statement` Paradromics. "Paradromics Receives FDA Approval to Expand Connexus Brain-Computer Interface Clinical Study Recruitment Nationwide." Press release, 26 October 2026. https://paradromics.com/news/paradromics-receives-fda-approval-to-expand-connexus-brain-computer-interface-clinical-study-recruitment-nationwide/

[^evtoday2024]: `news` "Synchron's Brain-Computer Interface Evaluated in COMMAND Early Feasibility Study." *Endovascular Today*, 3 October 2024. https://evtoday.com/news/synchrons-brain-computer-interface-evaluated-in-command-early-feasibility-study {Reports company-announced results presented at the 2024 Congress of Neurological Surgeons.}

[^medtechdive2025]: `news` Taylor, N. P. "Synchron raises $200M to prepare for brain computer interface launch." *MedTech Dive*, 7 November 2025. https://www.medtechdive.com/news/synchron-funding-bci-200m/804977/

[^intent2026]: `trial` Synchron, Inc. "INdependence Through Endovascular Neuroprosthetic Technology (INTENT): an Early Feasibility Study." ClinicalTrials.gov NCT07543367. https://clinicaltrials.gov/study/NCT07543367

[^reuters2026]: `news` Santhosh, C. "Elon Musk's Neuralink says it has 21 participants enrolled in trials." Reuters, 28 January 2026. https://finance.yahoo.com/news/elon-musks-neuralink-says-21-182925000.html