When might brain-computer interfaces for ALS be available outside a clinical trial?

BrainGate2 is an investigational device; no cleared commercial product exists. A path to regulatory approval would require expanded trials involving larger participant groups. The FDA has not yet established a formal approval pathway for this class of device. Publication in Nature Medicine marks a significant clinical milestone, but not the start of a commercial timeline.

Casey Harrell lost his speech to ALS — a brain implant returned it for 3,800 hours

Casey Harrell, 47, lost intelligible speech and the use of his arms and legs to ALS. What the neurodegenerative disease could not reach was the motor cortex signal — the pattern of neural activity his brain still generates when he tries to speak. A system of four microelectrode arrays, surgically placed in the left precentral gyrus and recording from 256 cortical electrodes, reads those signals and converts them into text and cursor control. Harrell communicated more than 183,000 sentences, close to two million words, at speeds that reached 56 words per minute.

The result matters because of where it happened. Previous brain-computer interfaces had achieved impressive accuracy in controlled research settings, but required a researcher to be present for each session to manage calibration and verify that the implant was recording correctly. Harrell used this system at home, on his own, on a near-daily basis, without that support. The device functioned for 3,800 hours.

That number is not a performance ceiling from a laboratory test. It is what independent daily use adds up to over nearly two years.

How the system works

The BrainGate2 device decodes two simultaneous streams of information. The first is an intended-speech signal: when Harrell attempts to say a word, neurons in his left precentral gyrus — the cortical region that coordinates speech movements — generate a recognizable pattern the system translates into text. The second is a cursor signal: decoded intended arm movements give him control of a standard computer interface.

The decoding algorithms are trained during initial calibration sessions. Once trained, they require no daily setup and no researcher to verify that the electrodes are recording cleanly. Harrell ran the system himself, on his own hardware. The research team sent periodic software updates as they refined their models, but the day-to-day operation was entirely his.

In laboratory testing with a 125,000-word vocabulary, word-level accuracy reached 99 percent. In everyday home use — where the vocabulary is unconstrained and the environment is not optimized — Harrell rated 92 percent of his sentences as accurate or mostly correct. Both figures exceed the accuracy standards used for clinical augmentative communication devices.

What the numbers actually show

The previous benchmark for this kind of system was 97 percent word accuracy in a controlled setting, achieved with a researcher present. This result moves both metrics forward simultaneously: higher accuracy, no researcher required.

At 56 words per minute, Harrell communicates at roughly twice the speed of the fastest existing eye-tracking augmentative communication systems. ALS is a progressive disease; as it advances, even residual eye movement can become unreliable. A cortical BCI bypasses that vulnerability by reading the signal before it travels to any muscle.

The 2 million words Harrell communicated over the study period include messages to colleagues, conversations with family, support for his continued employment, and exchanges that would otherwise have required slower, more laborious alternatives. “It is a life that is more full of dynamic action and with friends and family, colleagues, allowing me to communicate more naturally than any other technology experienced,” he said.

“BCIs have been proof-of-concept devices in research labs,” said David Brandman, the UC Davis neurosurgeon and co-principal investigator. “This work shows we may have crossed a threshold, empowering a person with paralysis to speak on his own terms.”

What this does not yet settle

The BrainGate2 trial is a small-participant research study. Harrell’s results reflect one person’s experience with one implant configuration over one specific window of time. ALS is a progressive condition, and the paper does not report how the quality of the neural signal changed as his disease advanced — an important variable for understanding how long any given patient might expect stable performance.

The hardware itself requires open neurosurgery to implant. The electrode arrays sit in the cortex; wires exit through the skull to a percutaneous connector. No fully wireless, fully internalized version of this class of device has yet reached clinical trials. Surgical access, cost, and post-operative care requirements mean that even a commercially approved version would reach only a fraction of the ALS population initially.

“We’ve made improvements bringing this medical technology closer to clinical usefulness,” said Sergey Stavisky, co-senior author and UC Davis neuroscientist. “He can use it at home without researcher support, achieving 99% accuracy while keeping up with faster speech attempts.” What remains is the path from closer to there — regulatory approval, manufacturing scale, and insurance frameworks for BCIs that do not yet exist in any jurisdiction.

Common questions about brain implants for ALS

What is ALS, and why does it take away speech?

ALS (amyotrophic lateral sclerosis) destroys the motor neurons that carry signals from the brain to muscles. As the muscles controlling the face, tongue, and throat weaken, speech becomes unintelligible. The neurons in the motor cortex that encode the intention to speak are spared until the later stages of the disease — which is precisely where a cortical BCI can intercept the signal before it reaches the damaged motor pathway.

How does this differ from the eye-trackers and muscle sensors ALS patients currently use?

Standard augmentative communication devices depend on preserved voluntary movement — typically eye tracking or residual muscle contractions. As ALS progresses, those pathways degrade too. A cortical BCI reads the signal one stage earlier, before any muscle activity is required, which means it can function in people for whom conventional devices have already begun to fail.

Does the implant need to be reset or recalibrated each day?

The study describes near-daily independent home use without researcher involvement, with only periodic software updates from the team. The core decoding model, once trained, did not require daily recalibration. That stability over nearly two years is one of the study’s primary findings.

When might this be available outside a clinical trial?

BrainGate2 is an investigational device; no cleared commercial product exists. A path to regulatory approval would require expanded trials involving larger participant groups. Publication in Nature Medicine marks a significant clinical milestone, but not the start of a commercial timeline.

BrainGate2 is currently enrolling additional participants. The research team’s next phase will attempt to replicate the results across a larger cohort, measure signal stability over longer time horizons, and determine whether further software improvements can push communication speed beyond 56 words per minute toward normal conversation rates.

Reference: Card et al., “Independent and accurate at-home use of a speech BCI by a participant with ALS,” Nature Medicine, 2026. DOI: 10.1038/s41591-026-04414-6

Tags: neuroscience, ALS, brain-computer interface, BrainGate, Casey Harrell, nature medicine