what a brain-computer interface can do for ALS

Scope: BCI companies, clinical work, research, funding, policy and neural data activity first reported or materially changed from 11 to 17 August 2026.

It was a quiet week for hardware and a loud one for the scaffolding around it. No new device launched and no regulator ruled. Yet the two things that decide whether any of this reaches a patient both moved. California advanced two neural data bills. A Chinese state insurer wrote the first policy for brain implant surgery. So this issue steps back from the news and asks the plain question a family asks after the diagnosis. What can a brain-computer interface actually do for someone with ALS right now.

The technology is real. It reaches almost no one, and you cannot buy it.

You have seen the headline. A man with ALS, unable to speak, talking again through wires in his brain. It is real. In the BrainGate2 trial he used an intracortical speech implant at home for more than 3,800 hours across about 19 months, produced over 183,000 sentences, and had 92 percent of them judged at least mostly correct.¹ An earlier participant with ALS reached 62 words a minute.² This is the strongest result in the whole field, and it belongs to a person with ALS.

We have shown you that result before, so this issue is not that story. Nothing new about ALS interfaces was published this week. Instead this is everything the headline leaves out, because that is the part a family needs and the part almost nobody prints.

Start with the number. Only a few dozen people with ALS in the entire published record are known to have received an implanted BCI.³ Every figure you have read is one, two or four people. The 56 words a minute belong to one man. These are extraordinary demonstrations. They are not yet a treatment, and honest reporting holds both facts at once.

It is also for a narrower moment than it sounds. For a person whose eyes still work, an eye-gaze keyboard is faster than any BCI and needs no surgery, 15 to 30 words a minute with a couple of minutes of setup.⁴ A BCI earns its place where gaze fails, and gaze does fail. It becomes unreliable in an estimated 15 to 35 percent of people with ALS from eyelid weakness, dry eye or loss of fixation, and in late disease a commercial eye tracker can stop working entirely.⁵ That is the real indication. Not faster than a working eye tracker. Present when the eye tracker is gone.

The common belief is that patients abandon this technology. For ALS the evidence says the opposite. Acceptance of communication technology in ALS runs around 96 percent.⁶ The real barrier is not rejection. It is getting a device at all. In one group only 61 percent of prescribed communication devices were ever delivered, lost to insurance denials, delays and the speed of the disease.⁶ The failure is in the system around the patient, not the patient's willingness.

Here is the range that has actually been shown in people with ALS, graded plainly. Intracortical typing has run at 13 to 32 correct characters a minute.⁷ The endovascular Stentrode, placed through a vein without opening the skull, gave four people with ALS or the related condition PLS twelve safe months and click-based control, though the eyes still moved the cursor and the implant only supplied the click.⁸ A non-invasive visual speller reached about 79 percent accuracy in ten people with ALS.⁹ A P300 system was used at home by one person for more than two and a half years.¹⁰ A simple brain switch let three people with severe ALS call a caregiver in about six seconds.¹¹ None of it is fast. All of it is real, and much of it needs no surgery.

The founding promise is the one still unmet. The reason the field exists is the completely locked-in patient, the person with no movement left. That is the one group the evidence has not reached. A single 2022 study reported spelling in one completely locked-in person with ALS at about one character a minute, and it stands unretracted. But its lineage is troubled. The earlier headline claims of communication in the completely locked-in state were retracted in 2019 after formal misconduct findings by the German research foundation.¹² Across all three of our independent research runs the honest grade for locked-in communication is the same. Claimed once, not independently confirmed.

The longest story we have runs seven years, and it ends the way ALS ends. A woman in the Netherlands used a fully implanted speller at home for seven years. Her use rose as her eye control failed, then declined as the disease advanced, and communication became unreliable when her neural signals weakened with progressing atrophy. The device still worked.¹³ On the single longest case in the record, the thing that ends an ALS interface is not the hardware. It is the disease outrunning the signal.

Every implanted communication BCI in the United States and Europe is investigational, reached only through a trial.¹⁴ China's approval of the Neuracle NEO, the world's first commercial invasive BCI, is real and does not cover ALS. It is cleared for grasp assistance in spinal cord injury.¹⁴ Inside a trial the device is free, but families still carry travel, caregiving and everything around it. There is no Medicare benefit category and no billing code for an implanted BCI. The covered path is eye-gaze and speech-generating devices, not the brain implant.¹⁵

The safety record, to be fair, is the reassuring part. Across 14 years and more than 12,000 implant-days the BrainGate trials reported six serious device-related events, no permanent neurological harm and no deaths.¹⁶

So the honest answer for a family is this. A BCI is a real option for a person with ALS today only inside a trial, only within reach of a specialised centre, only with a stable caregiver, and only for the few who qualify. At its documented best it gives months of near-conversational speech at home. What it cannot yet offer is a product you can buy, an insurer who will pay, a guarantee it will outlast the disease, or proven help for the completely locked-in. The advice every source converges on is unglamorous and sound. Plan communication early, bank your voice and use eye-gaze while they work, keep a low-tech fallback, and treat a BCI as research, not rescue.

1. California advanced both neural data bills past the point where bills usually die.

On 13 August the California Senate Appropriations Committee cleared AB 1883 and AB 1542 off its suspense file, the fiscal chokepoint where most bills quietly end. AB 1883, on workplace surveillance tools including neural data, passed as amended, 5 to 2. AB 1542, which prohibits the sale and sharing of sensitive personal information, a category California extended to neural data in 2024, passed 5 to 2. Both now go to the Senate floor before the session ends this month. Issue 7 reported these two bills stalled on this file. They are moving again. Neither is law yet.¹⁷

2. China wrote the first insurance policy for brain implant surgery.

On 12 August the state insurer PICC Property and Casualty issued what Chinese financial press describe as the country's first commercial insurance policy for the surgery to implant an invasive BCI, written with a hospital affiliated to Zhejiang University School of Medicine in Hangzhou. It covers surgical accidents, complications and malpractice liability during implantation. Set beside China's approval of the Neuracle NEO in March and its provincial service pricing, this is a country building approval, pricing and insurance in sequence. The West has none of the three.¹⁸

3. The measurement framework reimbursement will hang on was published.

A group spanning Mount Sinai, UMC Utrecht and Radboud published a unified framework of functional clinical outcomes for implantable motor BCIs, in Neurorehabilitation and Neural Repair on 15 August. Regulators and payers do not approve or reimburse on typing speed. They require validated measures of daily function. This paper proposes the shared endpoints that pivotal trials will need. It landed in the same week this deep dive documents how completely that pathway is still missing.¹⁹

4. BrainGate showed the brain's getting-ready signal can steer a cursor before movement.

In Current Biology, dated 11 August, the BrainGate group reported that preparatory activity in human motor cortex, recorded in three people with tetraplegia, is tuned to the direction of an intended movement before the movement is attempted, and that feeding this preparatory signal into the decoder gives measurably faster cursor control. It is a step toward control that anticipates intent rather than chasing it.²⁰

5. Two Parkinson's patients learned to drive their own brain stimulation.

A preprint from a UCSF-led group reports two people with Parkinson's disease who, after training at home with a simple game, learned to turn a cortical brain signal up or down at will and used it to raise or lower their own deep brain stimulation without moving a muscle. It is a closed loop where the patient, not a passive biomarker, is the controller. Two participants, no clinical endpoint yet.²¹

6. A single-neuron implant stayed stable for 1,240 days in a stroke-damaged brain.

A preprint reports one person with chronic aphasia after a large left-hemisphere stroke whose four implanted arrays held stable recording quality across 111 sessions over 1,240 days, matching results from people without brain lesions. Most long-term implant data comes from spinal cord injury or ALS. Showing it holds in a lesioned brain widens the group a speech implant might one day reach. One participant.²²

7. A 433-session human safety dataset for a new noninvasive stimulation method.

Researchers reporting for the company Magnetic Tides published a safety, tolerability and blinding study of kilohertz transcranial magnetic perturbation, across 433 sessions in 143 people including chronic stroke patients, with no device-related adverse reaction and successful blinding against a sham. It establishes safety, not benefit, and it sits on this issue's other theme, the many ways to put signal into the brain without opening the skull.²³

This issue combined three weekly research passes and three STORM evidence files, each produced by a separate AI model working from a written brief, then merged and verified against primary sources before publication. STORM is a multi-perspective research method from Stanford. Three of three independent support was treated as established for the merge. Two of three support could be published with the missing confirmation stated. One of three support required primary verification. Structured APIs were used for ClinicalTrials.gov, PubMed, Crossref and preprint discovery, and public pages and primary instruments were then opened for verification.

The value of running three models is not that they agree. It is that they fail in different directions and catch each other. This week one model produced a citation for a peer-reviewed trial result that did not exist. The other two had the correct record, and the merge removed it before publication. That is the method working as intended.

A note on scope. Synapse Dispatch reports on brain-computer interface developments already in the public domain. It is journalism and analysis, not laboratory or biological research, and it publishes nothing confidential. The value we add is verification and context, not new experiments.