The Voice in the Machine: Brain Implants Are Restoring Speech for ALS Patients — But Who Gets Access?

By July 2023, Casey Harrell could no longer speak to his four-year-old daughter. Amyotrophic lateral sclerosis had stolen his voice, progressively destroying the motor neurons that controlled his mouth, tongue, and larynx. That same month, neurosurgeon David Brandman at UC Davis implanted four microelectrode arrays into Harrell's left precentral gyrus — the brain region that coordinates speech production — and connected them to a system that would translate his neural activity into audible words [1].

Within minutes of activation, Harrell was communicating. Over the next 32 weeks and more than 248 hours of use, the system decoded his intended speech with up to 97% accuracy on controlled vocabulary tasks — the highest reported for any speech brain-computer interface (BCI) to date [2]. By 2025, a refined version described in Nature allowed him to speak through a synthesized voice at rates approaching natural conversation, with the ability to change intonation and even sing simple melodies [3].

The story is remarkable. It is also, at present, a story about one man.

How Fast the Field Is Moving

The speed of improvement in BCI speech decoding has been striking. In 2021, the best systems managed roughly 15 words per minute. By 2023, a Stanford team led by Francis Willett published results in Nature showing 62 words per minute — a 3.4x leap — with a 23.8% word error rate on a 125,000-word vocabulary [4]. A 2024 study in the New England Journal of Medicine pushed that to 78 words per minute [5]. The UC Davis system used by Harrell reportedly exceeded 97 words per minute in some sessions [2].

Source: Nature, NEJM, published BCI studies

Data as of Jun 1, 2025CSV

For context, typical conversational speech runs 150–160 words per minute. The gap is closing, but it remains real. And lab performance does not automatically transfer to the messiness of daily life. In Harrell's case, listeners could understand about 60% of synthesized words correctly during open conversation — compared to 4% without the BCI — which represents a dramatic improvement but also underscores how far the technology sits from transparent communication [1].

Academic research output reflects the acceleration. More than 54,000 papers have been published on BCI speech topics since 2011, with output peaking at over 7,500 papers in 2024 [6].

Source: OpenAlex

Data as of Jan 1, 2026CSV

The Competitive Landscape: Who Is Building What

At least five organizations are running or preparing human trials for implantable BCIs, each with different technical approaches and regulatory timelines.

BrainGate, the academic consortium behind Harrell's implant, has the longest track record, with active trials at UC Davis, Stanford, and Brown University focused on both speech and motor restoration. Their system uses Utah microelectrode arrays manufactured by Blackrock Neurotech, and remains the only platform with published peer-reviewed data on high-accuracy speech decoding [2][4].

Neuralink, founded by Elon Musk, received FDA approval for its PRIME study in May 2023 and has enrolled 21 participants as of early 2026. The company has expanded trials to the UK with its GB-PRIME study. Its primary focus has been motor control — allowing paralyzed users to operate computers and phones — rather than speech specifically [7].

Paradromics became the first company to receive FDA Investigational Device Exemption (IDE) approval for a fully implantable BCI specifically intended for speech restoration, with its Connexus device cleared in late 2025. The Connect-One study will initially enroll just two participants [8]. Paradromics claims data transfer rates of over 200 bits per second — 10 to 20 times higher than demonstrated in Neuralink's trials [9].

Synchron takes a less invasive approach, threading a stent-mounted electrode (the Stentrode) through blood vessels to reach the brain without open surgery. It has enrolled ALS patients since 2022, backed by investors including Jeff Bezos and Bill Gates [10]. The tradeoff: endovascular approaches record from fewer neurons at lower resolution.

Precision Neuroscience is conducting early feasibility studies with a thin-film electrode array placed on the brain surface, requiring only a small slit in the skull rather than full craniotomy [11].

The Trial Numbers Problem

Across all these programs, the total number of people who have received speech-restoring BCIs remains vanishingly small — likely fewer than 20 worldwide, with the BrainGate consortium accounting for the majority. Paradromics plans to start with two. Neuralink's 21 participants are primarily focused on motor tasks [7][8].

The BrainGate trial enrolled 14 participants over multiple years, with an average implantation duration of 872 days [12]. These are not the numbers of a technology approaching broad clinical deployment.

The path from investigational device to FDA-approved standard of care typically requires Phase I feasibility studies, then pivotal trials with larger cohorts, followed by a Premarket Approval (PMA) application. For a technology this novel, that timeline is measured in years, not months. No BCI company has yet begun a pivotal trial for speech restoration. A conservative estimate puts FDA approval for a commercial speech BCI at 2030 at the earliest, and that assumes no major setbacks [9][11].

Surgical Risks and Comparisons

Implanting microelectrode arrays requires craniotomy — opening the skull — and carries the standard neurosurgical risks: infection, bleeding, seizures, and delayed recovery. The BrainGate trial reported 68 device-related adverse events across 14 participants, including 6 classified as serious. However, none required device removal, and no participants developed brain infections or experienced permanently increased disability. The most common complaint was skin irritation at the percutaneous connector site [12].

How does this compare to deep brain stimulation (DBS), the most widely implanted neurological device? DBS infection rates range from 1.2% to 23% depending on the study, with a commonly cited aggregate figure around 3–9% [13][14]. Battery failures occur in roughly 8.4% of DBS cases, and lead migration in about 8.6% [15]. The BrainGate safety profile appears favorable by comparison, though the sample size is far too small for confident generalization.

A critical difference: DBS systems are fully implanted with no external connector, while most current speech BCIs (except Paradromics' and Neuralink's designs) still use percutaneous connections — wires passing through the skin — which create a persistent infection pathway. Fully implantable systems should reduce this risk but introduce their own engineering constraints around wireless data transmission and power [8][9].

What It Actually Costs

In clinical trials, participants pay nothing. The research institutions and their funding sources — primarily NIH grants, DARPA contracts, and venture capital — absorb all costs [16].

But those costs are substantial. A BCI pricing analysis estimates the total system cost breaks down roughly as follows: the implant device itself accounts for 40–55% of the total, surgical procedure and hospital stay for 25–35%, programming and calibration for 10–15%, and long-term support and replacement for 5–10% [16].

Based on analogous neurotechnology — DBS systems at $35,000–$100,000, cochlear implants at $30,000–$50,000, and the NeuroPace responsive neurostimulation system at $55,000–$90,000 — the first commercial BCI implants are expected to cost $50,000 to $100,000 or more at launch [16]. That figure does not include the specialized computational infrastructure required to run real-time neural decoding algorithms, the ongoing software updates, or the clinical support teams that current research participants rely on.

By contrast, eye-tracking augmentative and alternative communication (AAC) devices cost $8,000–$15,000 and are already reimbursable. Medicare covers 80% of the cost of speech-generating devices for ALS patients, with supplemental insurance or Medicaid often covering the remainder [17]. Some states, like Texas, cover the full cost [17]. The BCI reimbursement pathway does not yet exist.

The Equity Gap

ALS itself is not evenly distributed. CDC data show the disease is most common among white males aged 60–69, with an overall U.S. incidence of 1.44 per 100,000 persons [18]. The CDC estimates approximately 33,000 Americans currently live with ALS, a number projected to exceed 36,000 by 2030 [19].

Source: CDC National ALS Registry

Data as of Jan 1, 2025CSV

BCI clinical trials have been concentrated at a handful of elite academic medical centers: UC Davis, Stanford, Brown/Massachusetts General, and Johns Hopkins [1][2][20]. Participants must live near these sites or relocate for extended periods, undergo extensive screening, and commit to frequent research sessions — requirements that functionally exclude patients in rural areas, those without financial reserves, and those whose disease has progressed to the point where travel is impossible.

The broader clinical trial literature documents well-established barriers to diverse enrollment: lack of awareness, transportation challenges, work schedule conflicts, mistrust of medical institutions, and narrow eligibility criteria [21]. BCI trials compound these issues by requiring neurosurgery and prolonged engagement with research teams, raising the bar for participation beyond what most patients can clear.

No published BCI speech trial has reported demographic breakdowns of its participants, making it impossible to assess how representative the enrolled population is of the broader ALS community.

The Resource Allocation Argument

For every Casey Harrell, there are thousands of ALS patients whose most pressing needs are earlier diagnosis, better symptom management, and access to palliative care infrastructure that many regions lack.

Neurologists and bioethicists have raised a legitimate question: at what point does the concentration of research funding on high-cost, high-visibility individual interventions come at the expense of investments that would benefit more patients?

The argument is not that BCIs are worthless — it is that the marginal dollar spent on BCI research may produce less total reduction in suffering than the same dollar spent on expanding multidisciplinary ALS clinics, funding disease-modifying drug trials, or improving palliative care access. NIH-funded palliative care grants increased 25% between the 2011–2015 and 2016–2020 periods, with funding rising 35%, but palliative care research still receives a fraction of total NIH neuroscience spending [22].

Defenders of BCI investment counter that the technology addresses a specific, devastating loss — the ability to communicate — that existing therapies cannot restore. They argue that the research generates fundamental neuroscience insights with applications far beyond ALS, and that cost curves for neural interfaces will follow the trajectory of other implantable electronics, dropping substantially as manufacturing scales [9][11].

Both positions have merit. The tension between them will intensify as BCI companies seek broader FDA approval and push for insurance reimbursement.

BCIs vs. Existing AAC Technology

The comparison to currently available communication aids is essential context.

Eye-tracking systems — such as those made by Tobii Dynavox — allow patients to select letters, words, or phrases by looking at a screen. They are non-invasive, commercially available, widely deployed, and insurance-reimbursable. They typically produce 10–20 words per minute [17][23].

Their limitation is dependence on functional eye movement. As ALS progresses, some patients lose reliable eye control, leaving them with no communication channel. This is the population for whom invasive BCIs offer the most compelling advantage: they work by recording directly from speech-related neurons, bypassing all muscular pathways [23].

Non-invasive EEG-based BCIs exist but produce far lower data rates — typically 1–5 words per minute — because signals recorded from the scalp are noisy and lack the spatial resolution of implanted electrodes [23]. They have not demonstrated anything approaching conversational-speed communication.

The practical picture: eye-tracking works well for most ALS patients and should remain the first-line communication tool. Invasive BCIs address the subset of patients for whom eye-tracking fails, but at dramatically higher cost, risk, and infrastructure requirements.

What Happens When the Company Disappears

The most unsettling question in the BCI field has already been answered — badly — by precedent.

In 2020, Second Sight, a California company that manufactured retinal implants, laid off most of its workforce and stopped supporting approximately 350 patients who had its devices implanted. Repairs and replacements became impossible. The company stopped responding to medical queries, including one patient who needed to know whether an MRI would be safe with the implant still in her head [24].

In 2013, NeuroVista, which made an experimental seizure-tracking brain implant, ran out of money. Trial participants were told to have their implants removed. One patient, who described the device as having transformed her life, tried to keep it — even offering to remortgage her house. She was the last person in the trial to have it removed, against her will [25].

These are not hypothetical scenarios. They are the documented human cost of building medical dependency on venture-funded startups operating in a regulatory environment that has no robust framework for post-market device abandonment.

Current FDA regulations require manufacturers to maintain records and report adverse events, but they do not compel a company to continue operating, provide technical support, or transfer device specifications to a successor entity. A patient whose BCI company goes bankrupt faces the prospect of carrying an unsupported implant with no avenue for repair, software updates, or safe removal coordination [24][25].

Neuralink, Paradromics, and Synchron are all privately funded companies. None is profitable. The question of what happens to their patients' implants in the event of corporate failure is not academic — it is the most foreseeable risk in the entire enterprise.

Where This Goes Next

The BCI speech field is moving faster than its regulatory, financial, and ethical infrastructure can keep pace with. The science is genuinely impressive: a technology that did not exist outside laboratory demonstrations five years ago now allows a man with ALS to hold real-time conversations with his family [1][3].

But the distance between a successful research demonstration and equitable clinical deployment is vast. Fewer than 20 people have received speech BCIs. No pivotal trial has begun. No reimbursement pathway exists. No legal framework protects patients from corporate abandonment. And the patients most likely to benefit — those who have lost all voluntary muscle control — are often the least able to access clinical trials.

The coming years will test whether BCI speech technology can become more than a proof of concept for the few. The trajectory of cochlear implants, which took roughly 20 years from first human trials in the 1970s to mainstream clinical adoption in the 1990s, offers a historical parallel — but cochlear implants did not depend on venture-funded startups or require the computational infrastructure of a small data center to operate [9].

For Casey Harrell and his daughter, the technology has already delivered something beyond measure. The challenge now is ensuring that what he received does not remain a privilege confined to the participants of a handful of clinical trials.