Decoding is not independence

Implanted brain–computer interfaces can now stream speech and recreate typing, but small cohorts and weak clinical outcome measurement define the real evidence gap.

Executive assessment

Brain–computer interfaces have crossed an important technical threshold. Implanted systems can decode attempted speech quickly enough to synthesize a voice, reconstruct intended typing, and control digital devices. These are genuine achievements. They are also frequently described with a word the current evidence cannot yet support: independence.

Independence is not a decoder score. It means that a person can use a complete system reliably, for a meaningful activity, in daily life, with acceptable surgical risk, calibration burden, caregiver support, device maintenance, and failure recovery. It also means that the function remains available over time—not only during a supervised research session.

This issue begins BCI Watch, a recurring Brief section focused on implantable clinical systems. The scope is deliberately narrow: human evidence, registered trials, regulatory decisions, surgical safety, durability, and outcomes that matter to patients. Consumer EEG, speculative enhancement, and company announcements without verifiable clinical evidence remain outside the boundary.

The next BCI milestone is not another record in bits per second. It is a reproducible gain in communication, control, or participation that survives outside the laboratory.

The field measures performance more often than benefit

Evidence: peer-reviewed systematic review; literature searched through 24 December 2024.

A 2025 systematic review identified 112 studies and 80 unique people who had used an implantable BCI for motor intent. Nearly half of the studies had been published since 2020, illustrating the acceleration of the field. But the outcomes remained dominated by engineering measures. Read the review.

Decoding metrics were reported in 69.6% of studies and task performance in 62.5%; only 17.9% assessed clinical outcomes. When clinical outcomes were included, they were heterogeneous. This makes it difficult to compare systems, aggregate results, or determine whether a technical gain changes how a person functions or feels.

The denominator also deserves attention. A field capable of producing highly visible demonstrations still has a clinical experience base measured in tens of people under the review’s definition. Results from one or two exceptional participants may establish feasibility, but they cannot resolve patient selection, cross-user generalization, long-term reliability, or the resources required for routine use.

Streaming speech is a major result—and still a single-participant result

Evidence: peer-reviewed implanted ECoG study in one participant with severe paralysis and anarthria.

In 2025, a high-density surface array over speech sensorimotor cortex drove a continuously streaming speech synthesizer. The system decoded neural activity in 80-millisecond increments and generated intelligible, fluent speech personalized to the participant’s pre-injury voice. Read the study.

The achievement matters because conversational delay changes the nature of communication. A system that produces correct text after a long pause is not equivalent to a voice that supports turn-taking. Streaming also creates a harder safety problem: the decoder must identify when speech begins and ends, avoid producing unintended output, and allow rapid correction without breaking the conversation.

The evidence boundary is equally important. This was one participant in a clinical trial. The study demonstrates a new technical and functional capability; it does not show population-level reliability, comparative effectiveness, reduced caregiver burden, or sustained independent use at home. Those are the outcomes required before “restored communication” can become a general clinical claim.

Attempted typing offers a familiar interface

Evidence: peer-reviewed technical report in two implanted BrainGate2 participants.

A 2026 intracortical neuroprosthesis decoded attempted bimanual QWERTY typing in two people with tetraplegia—one with amyotrophic lateral sclerosis and one with spinal cord injury. The fastest reported performance reached 110 characters, or 22 words, per minute with a word error rate of 1.6%. Read the study.

The interface is clinically interesting because it reuses a familiar motor representation instead of asking the user to learn an arbitrary control scheme. It also connects naturally to existing digital communication tools. But peak performance should not be confused with dependable daily performance. The next reports should include setup time, recalibration, fatigue, unassisted use, error recovery, availability across days, and the proportion of intended communication actually completed.

For an implanted BCI, surgery is part of the product

BCI discussions often separate the decoder from the implant. Clinically, the intervention includes the electrode, insertion method, craniotomy or endovascular access, connectors, power, data transmission, wound, imaging restrictions, revision strategy, and possible explantation. A better decoder cannot compensate for an unacceptable implant pathway.

A robot-assisted intracortical implantation report illustrates the surgical layer. Four arrays were inserted in one participant with tetraplegia after cervical spinal cord injury. The arrays were placed with a single insertion, recordings remained usable at 100 and 200 days, and the participant achieved 20 of 20 virtual arm-control attempts. A localized skin infection required debridement and antibiotics and resolved. Read the surgical report.

That complication is not a reason to dismiss the technology. It is a reason to report the complete intervention. BCI evidence should include haemorrhage, infection, wound and connector problems, vascular injury, seizure, device migration, signal degradation, reoperation, explantation, and restrictions on future MRI or treatment. These outcomes must be connected to the functional benefit obtained in return.

Trial watch: early feasibility remains the operative phrase

Evidence: ClinicalTrials.gov records reviewed 27 August 2026; statuses and planned enrolment can change.

Neuralink’s US PRIME study remains a recruiting, single-group early feasibility study with estimated enrolment of 15 people with tetraplegia or tetraparesis related to spinal cord injury or ALS. Its primary outcomes are device- and procedure-related adverse events at twelve months, with longer safety follow-up planned. View the registry record.

Synchron’s US COMMAND study lists six actual participants and is active but not recruiting. Its primary purpose is also device feasibility, with treatment-related serious adverse events assessed through twelve months. The endovascular approach changes the surgical route, but it does not remove the need to measure vessel patency, migration, antithrombotic burden, signal durability, and functional use. View the registry record.

These are not failed pivotal trials. They are appropriately small studies designed to learn whether a device and procedure are safe enough to develop further. The editorial error is to present early feasibility enrolment as if it were evidence of broad clinical effectiveness.

Regulatory watch: a cleared electrode is not a cleared restorative BCI

Evidence: official FDA 510(k) record; device-component clearance.

The FDA found Precision Neuroscience’s Layer 7-T substantially equivalent in March 2025 as a cortical electrode. Its authorized use is temporary recording, monitoring, and stimulation at the brain surface for up to 30 days. This is a meaningful regulatory milestone for high-density cortical interfaces, but it is not clearance of a complete, permanently implanted system that restores speech or digital independence. View the FDA record.

BCI Watch will distinguish the regulated object from the larger platform: electrode, insertion robot, implant, decoder, external computer, and effector may have different evidentiary and regulatory status. “FDA cleared” should never be allowed to migrate from one component to an unsupported claim about the whole system.

BCI Watch: what Neurosurgery.ai will count

BCI Watch will not attempt to cover every neurotechnology announcement. A development qualifies when it changes at least one clinically meaningful dimension:

Human evidence. A peer-reviewed implanted-human result with participant count, task, follow-up, and limitations.
Trial status. A new registration, first participant, completed cohort, posted results, protocol change, or safety signal documented in an official registry.
Regulatory status. An official decision with the exact device, indication, duration, and boundary of authorization.
Surgical evidence. Implantation, vascular access, wound, revision, explantation, imaging compatibility, or long-term interface safety.
Functional value. Communication, digital control, mobility, activities of daily living, quality of life, caregiver burden, or sustained independent use.

Each item will carry an evidence label. Company releases can identify a signal, but they will not establish efficacy. Consumer EEG, animal-only demonstrations, fundraising, and speculative enhancement claims will not enter the watch unless they materially change a clinical implant pathway.

What to build next

Create a BCI evidence card inside the Brief workflow. Record implant type, cortical target, indication, participant count, follow-up, functional endpoint, surgical events, trial identifier, regulatory status, and evidence tier.
Separate peak performance from availability. Report the best session, median performance, usable days, calibration time, failures, and unassisted use whenever the source permits.
Make patient function the top-level outcome. Track whether the system changes communication access, digital independence, activities of daily living, participation, quality of life, or caregiver burden.
Treat the surgical pathway as a first-class comparison. Intracortical, surface, and endovascular systems trade signal resolution against invasiveness, vascular risk, tissue response, removability, and long-term stability.
Delay a standalone BCI section. Publish high-signal items in the Brief first. Build a dedicated page only after the evidence set is large and stable enough to support structured comparison.

Bottom line

Implanted BCIs have moved beyond the question of whether neural intention can be decoded. The next question is whether a complete system can deliver a durable, meaningful function to different people in ordinary life at an acceptable surgical and operational cost.

neural signal → decoder performance → reliable device control → daily functional use → patient independence → durable clinical benefit

BCI Watch will follow that entire path. It will celebrate the technical breakthroughs, but it will reserve the strongest claims for the point at which performance becomes dependable human function.

Editorial note: evidence labels distinguish systematic review evidence, single- and two-participant peer-reviewed studies, clinical-trial registry records, a surgical report, and an FDA device record. This Brief is professional information only and is not clinical advice or clinical decision support.