How Brain-Computer Interfaces Actually Work—And What Neuralink Has Actually Shipped

Brain-computer interfaces have graduated from science fiction to clinical reality over the past decade—but the reality is narrower, more technically complex, and more consequential for specific medical applications than most popular coverage suggests. Understanding what BCIs actually are, how they work at the electrode level, and what Neuralink has actually shipped versus what it has announced requires separating the genuine advances from the promotional overlay.

What a Brain-Computer Interface Actually Is

A brain-computer interface is a system that creates a direct communication pathway between brain activity and an external device, bypassing the normal output channels (muscles, speech). The communication can be read-only (recording brain signals and translating them into commands), write-only (stimulating specific neural patterns to create sensations or suppress activity), or bidirectional. BCIs are not new—the first implanted electrode array to decode intended limb movements appeared in clinical research in the late 1990s—but the technology has improved substantially in resolution, miniaturisation, wireless capability, and decoding algorithms.

The fundamental neuroscience: neurons communicate through electrical impulses (action potentials). When you intend to move a hand, motor cortex neurons fire in patterns that correspond to that intended movement. Electrodes placed on or in the brain can detect these electrical signals. Enough electrodes, in the right locations, detecting enough individual neurons or local field potentials, can capture enough information about the intended movement to reconstruct it—not directly reading thought in a general sense, but detecting specific, localised neural patterns that correlate with specific intentions or perceptions.

Invasive vs Non-Invasive BCIs

The BCI field divides roughly between non-invasive systems and implanted (invasive) systems, with very different capability and risk profiles.

Non-invasive BCIs use electrodes on the scalp (EEG—electroencephalography) or other external sensors to detect brain activity without surgery. EEG is practical, affordable, and safe, but has poor spatial resolution—the skull and scalp tissue attenuate and diffuse the electrical signals, making it difficult to resolve individual neuron activity. EEG-based BCIs can detect broad patterns (attention state, imagined movement categories, emotional arousal) but cannot decode the fine-grained motor intention signals that would enable high-bandwidth communication or precise prosthetic control. Consumer EEG headbands (Emotiv, Muse) fall into this category—useful for meditation feedback and very coarse control interfaces, not capable of the dextrous prosthetic control depicted in demonstrations.

Invasive BCIs use electrodes implanted directly in brain tissue, providing orders of magnitude better signal quality than surface EEG. Penetrating electrode arrays—the Utah Array and more recent flexible polymer arrays—insert needle-like electrodes into the cortex to record from individual neurons. These systems can decode intended hand and finger movements with enough precision to control robotic arms, type on virtual keyboards, or move a cursor with accuracy comparable to a mouse. The trade-off is surgical implantation, infection risk, and the biocompatibility challenge of foreign material in brain tissue.

ECoG (electrocorticography) sits between these categories: electrode grids placed on the surface of the brain (under the skull, on the cortex) rather than penetrating it. Resolution is better than EEG, implantation less destructive than penetrating arrays, and several research groups have demonstrated high-bandwidth communication through ECoG with patients who have conditions preventing speech.

What the Research Has Actually Demonstrated

Academic BCI research has produced compelling demonstrations over the past decade. The BrainGate consortium has enabled paralysed patients to control robotic arms, type on computers, and communicate through imagined handwriting or speech—achieving communication rates of 90+ characters per minute through imagined handwriting in a 2021 Nature paper. Chang Lab at UCSF demonstrated synthesis of continuous speech from the neural activity of a paralysed patient attempting to speak, producing intelligible synthesised voice output from brain activity with significantly reduced error rates compared to earlier systems.

These demonstrations are real, meaningful, and peer-reviewed. They are also conducted in highly controlled research settings, with patients who have implants placed for clinical reasons, using complex external computing infrastructure that processes the neural signals. They represent the state of what’s possible under optimal conditions, not what’s deployable at scale.

What Neuralink Has Actually Shipped

Neuralink’s approach uses a flexible polymer thread electrode array with 1024 channels, inserted by a surgical robot (the “R1”) designed to precisely place the thin threads between cortical blood vessels to reduce bleeding and inflammation. The device (called “N1”) transmits data wirelessly and is powered inductively through the skull—no external wires penetrating the skin, which reduces infection risk compared to earlier research implants.

As of mid-2026, Neuralink has implanted its N1 device in several patients in its PRIME Study (Precise Robotically Implanted Brain-Computer Interface), which is its FDA-authorised clinical trial. The first patient, Noland Arbaugh—a quadriplegic—demonstrated playing chess, controlling a computer cursor, and browsing the web using the implant. Subsequent patients have also demonstrated cursor control and related functionality.

This is genuine progress: a fully wireless, fully implanted BCI that enables meaningful computer control for paralysed individuals, demonstrated in humans, with FDA-authorised trial status. It is also narrower than Neuralink’s promotional claims about eventual capabilities (seamless memory augmentation, AI symbiosis, high-bandwidth telepathic communication). The current clinical application is assistive technology for people with significant motor impairment—a genuinely important and valuable use case that is very different from the mass-market cognitive enhancement that Musk’s public statements describe.

The Challenges That Remain

Several technical challenges constrain what current BCIs can achieve and how widely they can be deployed.

Signal stability: penetrating electrodes cause tissue response (glial scarring) over months to years, which progressively insulates the electrode from nearby neurons and degrades signal quality. Long-term performance of implants—whether the R1’s gentler insertion technique substantially improves multi-year stability—remains an open question in Neuralink’s clinical data, which is too early to fully assess.

Decoding bandwidth: the brain’s information processing involves roughly 86 billion neurons with complex, population-level coding. 1024 channels records a tiny fraction of this. Current BCIs can decode motor intention for specific, well-studied cortical areas, but extending to more complex cognition—language comprehension, abstract thought, emotional state—requires recording from many more neurons distributed across much larger brain areas.

Ethics and access: BCIs will not initially be accessible to most people who might benefit from them, due to cost, surgical risk, and the complexity of clinical support. The regulatory, insurance, and equity frameworks for deploying these devices responsibly are nascent.

The honest assessment: BCIs are a real, advancing field with genuine near-term applications in assistive technology for people with severe motor impairments. The longer-term applications that dominate public discussion—enhancement, augmentation, AI integration for healthy individuals—require technical advances that are possible in principle but not near-term in practice. Neuralink’s current results are meaningful within that realistic frame.