Brain-computer interfaces (BCIs) sit at an unusual intersection of genuine medical achievement and heavy speculation. In the same year, a person who had lost the ability to speak regained near-conversational communication through electrodes implanted in their brain, and headlines promised telepathy, memory uploads, and consumer "neural" gadgets that read your mind. Sorting the demonstrated from the hypothetical matters, because the gap between the two is wide and consequential. This article explains what a BCI actually is, the spectrum of hardware from scalp sensors to implanted arrays, what peer-reviewed clinical work has genuinely shown, the engineering limits that constrain the field, and the privacy, safety, and regulatory questions that are still being worked out.
What a Brain-Computer Interface Actually Is
A brain-computer interface is a system that records signals associated with brain activity, decodes them with software, and translates the result into commands for an external device such as a computer cursor, a robotic arm, or a speech synthesizer. The U.S. National Institute of Neurological Disorders and Stroke describes these technologies as a means of restoring function for people with paralysis and other neurological conditions by reading neural activity and acting on it [1]. Critically, a BCI does not read "thoughts" in any general sense. It detects patterns of electrical or metabolic activity, usually tied to a specific intention such as trying to move a hand or trying to speak, and maps those patterns to outputs.
The core challenge is that the brain's signals are faint, noisy, and overlapping, and the relationship between a neural pattern and an intended action has to be learned by a decoding algorithm, often retrained over time. A BCI is therefore as much a machine-learning system as a piece of hardware, and its performance depends heavily on signal quality at the source.
The Spectrum: From EEG Headsets to Implanted Arrays
BCIs are usually classified by how physically close the sensors sit to brain tissue, which directly determines signal resolution and risk. The categories run roughly as follows:
- Non-invasive: Electrodes rest on or above the scalp. Electroencephalography (EEG) is the most common, measuring electrical activity through the skull; related methods include functional near-infrared spectroscopy (fNIRS) and magnetoencephalography (MEG). These require no surgery but capture a blurred, summed signal from large populations of neurons [2].
- Partially invasive: Electrodes are placed inside the skull but rest on the surface of the brain rather than penetrating it. Electrocorticography (ECoG) and endovascular arrays threaded through blood vessels fall here, offering better signal quality than scalp recording without entering brain tissue [2].
- Invasive: Microelectrode arrays are implanted directly into the cortex, recording from individual or small clusters of neurons with high spatial and temporal precision. This yields the richest data and powers most high-performance demonstrations, but carries surgical risk [2].
The trade-off is consistent across the field: more signal fidelity generally means more invasiveness, and consumer EEG products marketed for focus, meditation, or gaming sit at the lowest-resolution end of this spectrum. Their data is far coarser than anything used in the clinical results discussed below, a distinction that marketing often blurs.
What Has Actually Been Demonstrated
The most credible advances have come from invasive research systems used in tightly controlled clinical trials, and the results are real. In a study published in the New England Journal of Medicine, researchers working with the BrainGate2 trial implanted microelectrode arrays in a man with amyotrophic lateral sclerosis (ALS) who could no longer speak intelligibly. The system decoded his attempted speech from cortical activity, producing near-perfect accuracy on a constrained 50-word vocabulary and a word error rate of roughly 10 percent against a large vocabulary of about 125,000 words, with useful communication beginning on the first day of use [3]. Separately, intracortical BCIs have allowed people with paralysis to control computer cursors and robotic limbs by attempting movement.
Endovascular systems have also reached human trials. Synchron's Stentrode, a stent-mounted electrode array delivered through the jugular vein to rest against the motor cortex without open-brain surgery, was evaluated in the FDA-approved COMMAND early feasibility study, which reported that the device was deployed safely and could capture motor intent for controlling digital devices [4]. Earlier long-term safety results for the first-generation Stentrode, in a small group of participants with severe paralysis, were published in the peer-reviewed journal JAMA Neurology [4]. Neuralink's PRIME study has enrolled participants who use an implanted array to control computers, and its devices have received FDA breakthrough-device designations, a status that expedites review but is not the same as approval [5].
What has not been demonstrated is anything resembling the consumer mythology. No published, peer-reviewed work shows a BCI reading abstract thoughts, transmitting complete sentences between brains as a routine capability, writing memories into the brain, or restoring cognition in healthy people. The verified achievements are specific: decoding attempted movement or attempted speech in people with motor or speech impairment, under research conditions.
The Main Players and Approaches
A handful of organizations dominate the clinical conversation, each betting on a different point along the invasiveness spectrum. Described neutrally and factually:
- BrainGate is a long-running academic research consortium that has used penetrating intracortical microelectrode arrays for cursor control, robotic limb control, and the speech decoding described above [3].
- Synchron uses the endovascular Stentrode, prioritizing a less invasive delivery method at the cost of recording from fewer, less precise electrodes [4].
- Neuralink develops a fully implanted device with flexible electrode threads inserted by a surgical robot, aiming for higher channel counts and a wireless, cosmetically hidden implant [5].
- Precision Neuroscience pursues a thin-film surface array that sits on the cortex without penetrating it; the company reported pilot human data in Nature Biomedical Engineering in 2025 [6].
These approaches are not interchangeable. A penetrating array can resolve activity that a surface or endovascular device cannot, while a less invasive device is easier to deliver and potentially safer. There is no consensus that one design will win; the field is still establishing which level of invasiveness justifies which clinical benefit.
The Hard Technical Limits
The constraints on BCIs are physical and biological, not merely matters of more engineering effort. They explain why progress is incremental rather than exponential.
- Bandwidth: Even high-channel implants sample a tiny fraction of the roughly 86 billion neurons in the brain. Non-invasive EEG, blurred by the skull, conveys far less information still, which is why scalp-based systems cannot match implant performance for complex control.
- Longevity and signal stability: Implanted electrodes degrade. Long-term human data published in the Journal of Neural Engineering show that intracortical arrays can record usable signals for years, but signal quality declines over time, and devices have a finite functional lifetime [7]. Decoders must be recalibrated as the recorded signals drift.
- Biocompatibility: The brain treats an implant as a foreign body. An inflammatory response involving glial cells begins at insertion and is followed by scarring around the electrodes, alongside material degradation, all of which can erode recording quality over months to years [7].
These are the reasons a laboratory demonstration does not translate quickly into a durable, maintenance-free medical product. An implant that performs brilliantly for months must still prove it performs adequately for years, ideally without repeat surgery.
The Real Privacy, Consent, and Security Risks
Neural data is unusually sensitive because it is generated involuntarily and can, in principle, reveal information a person never chose to disclose. The risks are concrete rather than science-fiction. Neural recordings could be exposed in a breach, repurposed by a manufacturer or third party, or used to infer states the user did not consent to share, and as wireless implants and consumer EEG devices proliferate, the attack surface grows. Standards bodies including the IEEE have begun developing terminology, reporting, and safety frameworks for neural interfaces, reflecting a recognized need for safeguards against unauthorized access to and manipulation of these systems [8].
The legal picture is catching up unevenly. In 2024, Colorado and California became the first U.S. states to amend their consumer privacy laws to treat neural data as a protected category of sensitive personal information, and several other states have since moved to follow [9]. Federal protections remain fragmented; health-privacy rules such as HIPAA apply mainly to traditional healthcare settings and may not cover neural data collected by consumer device makers [9]. Consent is a further complication: many trial participants have severe disabilities, raising careful questions about comprehension, autonomy, and what happens to an implant and its data if a company fails or a study ends. This article is general information and not legal or medical advice; anyone weighing participation in a trial should consult the study's investigators and an independent professional.
Regulatory Status
Implanted BCIs are regulated in the United States as medical devices. In May 2021, the FDA issued final guidance titled "Implanted Brain-Computer Interface (BCI) Devices for Patients with Paralysis or Amputation - Non-clinical Testing and Clinical Considerations," which sets out recommendations for the bench testing and clinical study design expected in investigational device exemption applications [10]. The guidance emphasizes durability testing matched to the intended implant duration and treats the supporting software as warranting close scrutiny [10].
In practice, current human use occurs under investigational device exemptions and early feasibility studies, not full marketing approval. Breakthrough-device designations granted to several programs speed regulatory interaction but do not certify that a device is safe and effective for general use. As of 2026, no implanted BCI has received full premarket approval for broad clinical use in the United States, and the devices remain investigational.
The Bottom Line
Brain-computer interfaces have moved from proof of concept to genuine, documented clinical results: people with paralysis controlling devices, and people who lost speech communicating again through decoded neural activity. Those achievements are narrow, hard-won, and confined to research settings, and they should be neither dismissed nor inflated. The honest summary is that BCIs are an early-stage medical technology with real promise for restoring function, bounded by stubborn limits in bandwidth, longevity, and biocompatibility, and accompanied by privacy and consent questions that regulation is only beginning to address. For now, the most useful posture toward both the breakthroughs and the hype is the same: read the trial data, note what was actually measured, and treat everything beyond it as unproven.
Sources
[1] NIH National Institute of Neurological Disorders and Stroke: Brain-Computer Interfaces — https://www.ninds.nih.gov/health-information/disorders/brain-computer-interfaces
[2] National Academies / NCBI Bookshelf: Brain-Machine and Related Neural Interface Technologies — https://www.ncbi.nlm.nih.gov/books/NBK588597/
[3] New England Journal of Medicine: An Accurate and Rapidly Calibrating Speech Neuroprosthesis — https://www.nejm.org/doi/full/10.1056/NEJMoa2314132
[4] Synchron: Positive Results from the U.S. COMMAND Study of an Endovascular Brain-Computer Interface — https://www.businesswire.com/news/home/20240930433219/en/Synchron-Announces-Positive-Results-from-U.S.-COMMAND-Study-of-Endovascular-Brain-Computer-Interface
[5] FDA: Breakthrough Devices Program — https://www.fda.gov/medical-devices/how-study-and-market-your-device/breakthrough-devices-program
[6] Nature Biomedical Engineering: Minimally Invasive Implantation of Scalable High-Density Cortical Microelectrode Arrays (Precision Neuroscience) — https://www.nature.com/articles/s41551-025-01501-w
[7] Journal of Neural Engineering (IOP): Long-term intracortical microelectrode array performance in a human — a 5-year retrospective analysis — https://iopscience.iop.org/article/10.1088/1741-2552/ac1add
[8] IEEE Brain: Standards for Neurotechnology and Brain-Computer Interfaces — https://brain.ieee.org/resources/standards/
[9] Arnold & Porter: Neural Data Privacy Regulation — What Laws Exist and What Is Anticipated — https://www.arnoldporter.com/en/perspectives/advisories/2025/07/neural-data-privacy-regulation
[10] FDA: Implanted Brain-Computer Interface (BCI) Devices for Patients with Paralysis or Amputation — Non-clinical Testing and Clinical Considerations (Final Guidance, 2021) — https://www.fda.gov/regulatory-information/search-fda-guidance-documents/implanted-brain-computer-interface-bci-devices-patients-paralysis-or-amputation-non-clinical-testing
