Brain-computer interfaces are crossing from neuroscience demonstrations into clinical engineering. People with paralysis have used implanted systems to move computer cursors, select letters, and turn attempted speech into text or synthesized audio. Companies are recruiting participants for early feasibility studies, and regulators have published guidance for evaluating implanted devices. The field is leaving the laboratory, but it is not leaving experimentation behind.

Communication is the clearest near-term application

A brain-computer interface, or BCI, measures neural activity and translates patterns into an external command. Noninvasive systems can use electroencephalography or imaging, while implanted systems record closer to the neural source. Implantation can improve signal quality and bandwidth, but introduces surgical risk, long-term device maintenance, and a demanding regulatory pathway.

Recent speech studies show why invasive approaches attract attention. In 2023, a Stanford-led team reported a participant with amyotrophic lateral sclerosis using four intracortical arrays to decode attempted speech at 62 words per minute. The system achieved a 23.8% word error rate with a 125,000-word vocabulary (Nature). A separate University of California, San Francisco team used a surface electrode array to decode text at a median 78 words per minute and animate speech-related movements on a digital avatar (Nature). Each result came from one participant, a critical limitation often lost in broad claims about mind reading.

In 2024, BrainGate researchers reported a system used by a 45-year-old man with ALS. It reached 99.6% accuracy on a 50-word vocabulary on its first day of use and 90.2% accuracy after expanding beyond 125,000 words on the second. With more training, it sustained 97.5% accuracy over 8.4 months, and the participant used it in self-paced conversation for more than 248 cumulative hours at roughly 32 words per minute (New England Journal of Medicine). The longitudinal use is at least as important as the headline accuracy because assistive technology must work repeatedly outside a single benchmark session.

Trials are testing different routes into the brain

The U.S. Food and Drug Administration issued final guidance in 2021 for implanted BCIs intended for people with paralysis or amputation. It described the field as “progressing rapidly from fundamental neuroscience discoveries to translational applications” and set recommendations for nonclinical testing and early and pivotal study design (FDA). That language is forward-looking, not an approval of a consumer product.

Neuralink's PRIME study, ClinicalTrials.gov identifier NCT06429735, is an early feasibility study of a skull-mounted wireless implant and a robot that inserts electrode threads. The company says its first participant received an implant in January 2024 and that multiple participants with ALS or spinal-cord injury have since received devices (Neuralink). The study is designed to evaluate safety and initial device functionality, not to prove population-wide effectiveness.

Synchron is pursuing an endovascular approach. Its Stentrode electrode array is delivered through the jugular vein into a blood vessel near motor cortex, avoiding open-brain surgery. A first-in-human study described two participants using the system for computer-based activities of daily living, with signals transmitted to a tablet and translated into click commands (Journal of NeuroInterventional Surgery). The procedure may reduce some surgical burdens, while placement through a vessel imposes different constraints on signal location and resolution.

A demo is not a durable clinical outcome

Brain-Computer Interfaces Are Leaving the Laboratory — supporting image
Brain-Computer Interfaces Are Leaving the Laboratory — supporting image

BCI performance depends on electrode stability, signal drift, decoding software, user training, fatigue, disease progression, and the task being measured. A vocabulary-constrained accuracy score cannot be compared directly with open conversation. Words per minute do not describe setup time, calibration, caregiver support, or failures. Small early studies are appropriate for feasibility, but they cannot reveal rare adverse events or performance across diverse patients.

The device lifecycle is unusually consequential. Participants need plans for software updates, battery and hardware failures, cybersecurity incidents, company closure, explantation, and years of follow-up. An implant may continue collecting sensitive neural signals after a trial ends. Consent must cover what is recorded, what derived data mean, who can access them, and whether recordings can be reused to train future models.

Privacy discussions should also resist exaggeration. Today's motor and speech BCIs decode activity associated with carefully trained tasks; they do not provide general access to a person's private thoughts. Yet inferred intention, attempted speech, and behavioral patterns are deeply sensitive. Security and user control must be designed before bandwidth rises, not added after commercialization.

Clinical usefulness should define progress

The most credible near-term goal is restoring agency to people with severe communication or motor impairments. Success should be measured by reliable communication, reduced caregiver dependence, patient-reported benefit, manageable risk, and continued support—not by the spectacle of moving a cursor with thought.

Brain-computer interfaces have left the laboratory in the limited but important sense that real participants are using them under regulated protocols for meaningful tasks. The next transition is harder: from exceptional individual results to devices that clinicians can prescribe, health systems can support, and patients can rely on for years.

Sources