Neuralink & Brain-Computer Interfaces: Complete Guide 2026

1. What Is a Brain-Computer Interface?

A brain-computer interface (BCI) is a system that creates a direct communication pathway between the electrical activity of the brain and an external device — bypassing the normal biological pathways (nerves, muscles, speech) that the brain uses to interact with the world.

In its most basic form, a BCI involves three components:

1. Signal acquisition: Capturing electrical signals from neurons, either at the scalp surface (non-invasive), from inside the skull but outside the brain tissue (epidural), or directly from neurons within the cortex (intracortical).
2. Signal processing: Filtering, amplifying, and decoding the captured neural signals to extract meaningful information — distinguishing intentional movement-related signals from background neural noise.
3. Output: Translating the decoded neural signal into a command for an external device — moving a cursor, typing a character, controlling a robotic arm, or in the future, sending a signal to another person's brain.

2. The Neuroscience: How Neurons Generate Signals

The brain contains approximately 86 billion neurons. Each neuron communicates by generating electrical impulses called action potentials (or "spikes") — brief (~1 millisecond) voltage changes of approximately 70–100 mV that travel along axons and trigger neurotransmitter release at synapses.

2.1 Local Field Potentials and Single-Unit Activity

BCIs can record two types of signals depending on electrode proximity to neurons:

• Single-unit activity (SUA): Individual action potentials from one or a few neurons, captured by microelectrodes placed within ~50 µm of the neuron soma. Highest information density. Requires intracortical implants (invasive). Neuralink's N1 chip targets this.
• Local field potentials (LFPs): Summed electrical activity of many neurons in a surrounding region (~1 mm radius). Can be recorded with larger electrodes further from neurons. Less precise but more stable over time.

2.2 Motor Cortex and Population Coding

The primary motor cortex (M1) contains neurons whose firing rates correlate with intended movement direction and force. A key finding: intended arm movements can be decoded from the activity of roughly 100–200 M1 neurons using relatively simple linear decoders. This is the computational foundation on which all motor BCI systems are built — including Neuralink's.

The population code is high-dimensional and redundant: many neurons contribute to each movement command, so losing some electrode contacts over time (due to neural drift or tissue response) degrades performance gradually rather than catastrophically.

3. Types of BCIs: Invasive vs. Non-Invasive

4. Neuralink: The Company and Its Mission

Neuralink was co-founded by Elon Musk and Max Hodak in 2016, along with seven neuroscience and engineering colleagues. Its stated mission is to "create a generalized brain interface to restore autonomy to those with unmet medical needs today and to unlock human potential tomorrow."

The company distinguishes itself from prior neural interface companies through its emphasis on miniaturization, wireless transmission, and surgeon-independent deployment. Earlier BCI systems (notably BrainGate) required large transcutaneous connectors that passed through the skull — cables connecting the implant to external hardware — increasing infection risk and severely limiting patient mobility. Neuralink's design is fully implanted and wireless.

Neuralink has raised approximately $700 million in funding (as of early 2026) and employs over 400 people across neuroscience, robotics, material science, and software engineering.

5. The N1 Chip: Technical Specifications

The N1 implant is a coin-sized (~23mm diameter) device implanted flush with the skull:

5.1 The R1 Surgical Robot

Thread insertion is performed by Neuralink's proprietary R1 robot, which uses computer vision and precision mechanics to insert the 64 electrode threads into specific cortical regions with micron-level accuracy, avoiding surface blood vessels. The surgical procedure takes approximately 25 minutes. The robot's precision is essential — human surgeons cannot consistently achieve the sub-millimeter accuracy required to thread 64 electrodes around a complex vascular network without causing bleed damage.

6. The PRIME Study: First Human Implants

The Precise Robotically Implanted BCI (PRIME) Study is Neuralink's first-in-human clinical trial, granted FDA Breakthrough Device Designation in 2023 and approved for human trials in May 2023.

6.1 Trial Design

PRIME is a clinical feasibility study with N=6 participants (as of Q1 2026). Inclusion criteria require participants to have quadriplegia or tetraplegia due to cervical spinal cord injury or ALS (amyotrophic lateral sclerosis) — above the level of C4 (which controls arm movement). The study's primary endpoint is the safety and feasibility of the implant; secondary endpoints measure BCI performance (cursor speed, typing rate, functional task completion).

6.2 Implant Sites

All PRIME study implants are placed in the hand knob area of the primary motor cortex — the region controlling hand and finger movements, where population coding of intended movements is best understood and most reliably decodable. This maximizes the signal available for cursor control (which requires the same neural representation as intended hand movement).

7. Noland Arbaugh & the Published Results

Noland Arbaugh, PRIME participant #1 (P1), became quadriplegic at age 22 following a diving accident causing a C4/C5 spinal cord injury. He received the N1 implant on January 29, 2024. Neuralink published detailed results in a peer-reviewed paper in Science in September 2024, co-authored by Neuralink researchers and independent academic collaborators.

7.1 Performance Metrics (Published)

• Peak cursor speed: 8.0 bits per second (BPS) — a primary BCI performance metric combining speed and accuracy. This exceeded any previously reported BCI result by approximately 2×.
• Typing speed via cursor: ~40 words per minute using point-and-click on an on-screen keyboard — approaching able-bodied typing speeds for many tasks.
• Daily BCI use: Noland reported using the system for 8+ hours per day — gaming, web browsing, social media, video calls — demonstrating practical, unsupervised use outside clinical settings.
• Latency: <10 ms from neural signal to cursor movement — imperceptible to the user.

7.2 Thread Retraction Issue

In April 2024, Neuralink disclosed that a subset of the 64 implanted threads in P1's device had retracted from the cortex, reducing the number of active recording channels. This is a known issue with intracortical flexible electrodes — the brain tissue exhibits a foreign body response (glial scarring) around implanted materials, which can cause chronic electrode displacement. Despite losing channels, P1's performance remained high because the remaining active channels still provided sufficient signal. Neuralink updated its decoding algorithm to compensate for the reduced channel count, actually improving performance metrics after the thread retraction through better signal processing.

7.3 Subsequent Participants

A second PRIME participant (P2), implanted in mid-2024, achieved higher initial channel counts using an updated thread array design that included chemical modifications to reduce the foreign body response. A third participant focused on a different use case: robotic arm control rather than cursor movement, providing initial data on bidirectional BCI (stimulation as well as recording). As of early 2026, four participants total have been implanted under the PRIME protocol.

8. Competitors: Synchron, BrainGate, Paradromics

8.1 Synchron — Stentrode

Synchron is arguably Neuralink's most clinically advanced competitor, having performed more human implants as of 2026. The Stentrode is an endovascular BCI — it is deployed via catheter through the jugular vein and positioned in the superior sagittal sinus (a large vein running along the top of the brain). No craniotomy is required.

The Stentrode expands like a stent to press against the blood vessel wall, recording cortical field potentials through the vessel. While signal quality is lower than direct intracortical recording (due to the blood vessel wall intervening), this approach dramatically reduces surgical risk and recovery time. As of early 2026, 12 patients in Australia and the US have received Stentrode implants, with demonstrated typing using thought-controlled software keyboards.

Synchron completed a 6-patient US feasibility trial with data published in the Journal of NeuroEngineering and Rehabilitation (2024), showing sustained use for up to 2 years post-implant with no serious device-related adverse events.

8.2 BrainGate

BrainGate is a research consortium (Brown University, Massachusetts General Hospital, Stanford, Providence VA) pioneering intracortical BCIs since 2004 using the Utah Array — a 10×10 silicon electrode array with connectors that exited through the skull. BrainGate participants have demonstrated cursor control, robotic arm operation, and neural speech decoding. The consortium's research has produced foundational knowledge for the entire field, though it does not have a commercial pathway — its R&D feeds academic publication and clinical development by commercial spinouts including BrainGate2.

8.3 Blackrock Neurotech

Blackrock Neurotech manufactures the Utah Array used in BrainGate and in many other research BCI studies, as well as implanting its own clinical systems. It has performed more than 40 human intracortical BCI implants — more than any other company — in patients with ALS, spinal cord injury, and locked-in syndrome. Commercial but research-oriented, not yet operating at the consumer scale Neuralink is targeting.

8.4 Paradromics

Paradromics is developing high-channel-count BCIs (target: 1 million electrodes) using a different architecture than Neuralink — wire microelectrode bundles rather than flexible polymer threads. The company has performed preclinical work in non-human primates and is working toward human trials under a different FDA pathway than PRIME. Their approach targets higher bandwidth communication with the brain, aiming for the future applications (memory augmentation, high-bandwidth communication) rather than the immediate medical restoration applications.

9. Non-Invasive BCIs: EEG, fNIRS, and Commercial Devices

For the vast majority of users — anyone without a medical condition requiring an implant — non-invasive BCIs are the relevant technology today and for the foreseeable near term.

9.1 EEG-Based BCIs

Electroencephalography (EEG) measures electrical potentials at the scalp via conductive electrodes. The signal is a heavily filtered version of underlying neural activity — individual neuron spikes are invisible; only large-scale synchronized oscillations (alpha waves, beta waves, P300 event-related potentials) can be detected. EEG BCIs use these rhythms to construct communication channels:

• Motor imagery BCIs: Users mentally simulate hand/foot movements, generating distinct ERD (event-related desynchronization) patterns in mu (8–12 Hz) and beta (13–30 Hz) bands that a classifier decodes as left/right commands.
• P300-based communication: The P300 component (a positive EEG deflection ~300ms after a rare, attended stimulus) allows users to select letters from a matrix by counting flashes — used in augmentative and alternative communication (AAC) systems for locked-in patients.
• SSVEP (steady-state visually evoked potentials): Fixating on a flickering target at a specific frequency produces a strong EEG response at that frequency — high-accuracy but requires intact vision and concentration.

9.2 Commercial EEG Devices

• OpenBCI: Open-source EEG hardware targeted at developers and researchers. Popular for BCI prototyping, neurofeedback, and art installations. Cyton (8-channel) and Ganglion (4-channel) boards with Raspberry Pi compatibility.
• Emotiv EPOC X: 14-channel wireless EEG headset with an SDK and marketplace for consciousness-state monitoring, accessibility controls, and research applications.
• Muse Headband: Consumer EEG meditation device (4 electrodes) that provides real-time feedback on mental states — primarily marketed for mindfulness, not BCI control.
• Neurosity Crown: Developer-focused EEG headset with an open SDK designed for integration with productivity software — notifications delivered when concentration drops, focus states triggering actions in software.

9.3 Kernel Flow (fNIRS)

Kernel, founded by Bryan Johnson, builds functional near-infrared spectroscopy (fNIRS) headsets — Kernel Flow — that measure cerebral blood oxygenation as a proxy for neural activity. While lower temporal resolution than EEG, fNIRS provides better spatial resolution and is not confounded by electrical muscle artifacts. Kernel has positioned Flow as a research-grade neuroimaging device for studying mental states, cognitive performance, and psychedelic treatment efficacy.

10. Medical Applications Beyond Motor Control

Motor restoration (cursor control, robotic arm operation) is the initial clinical application because the motor cortex is the best-understood brain region for population coding. But the broader medical opportunity for BCIs extends throughout neurology and psychiatry:

10.1 Neural Speech Decoding

Researchers at UC San Francisco (Chang Lab) and UC Davis demonstrated in Nature (2021, 2023) and New England Journal of Medicine (2023) that speech-related neural signals in the ventral sensorimotor cortex could be decoded into words at rates approaching natural speech — 62–80 words per minute in patients with paralysis affecting their voice. "Neural speech" BCIs read the intended speech motor plan, not acoustic output, enabling communication for patients who cannot speak — including ALS patients who have lost all motor function.

10.2 Memory Enhancement

Wake Forest Institute for Regenerative Medicine has demonstrated in human trials (published in the Journal of Neural Engineering) that hippocampal stimulation patterns — delivered by a BCI during the encoding phase of a memory task — can improve recall accuracy by 30–50% in patients with mild traumatic brain injury. This is not memory "reading" but targeted stimulation to enhance consolidation. DARPA's RAM (Restoring Active Memory) program funded much of this research.

10.3 Treatment-Resistant Depression and OCD

Deep Brain Stimulation (DBS), which delivers continuous electrical pulses to specific brain structures via permanently implanted electrodes, has FDA approval for Parkinson's disease, essential tremor, and dystonia, and a Humanitarian Device Exemption for OCD. Closed-loop DBS (detecting biomarkers of abnormal activity and delivering stimulation only when needed) is in trials for treatment-resistant depression — with early results showing substantial improvement in patients who failed all other treatments.

10.4 Epilepsy Management

NeuroPace RNS System is an FDA-approved closed-loop brain stimulator for drug-resistant focal epilepsy. It continuously monitors electrocorticographic activity, detects seizure onset patterns, and delivers counter-stimulation to abort seizures before they propagate. In a 10-year follow-up study published in Neurology, 73% of patients reported ≥50% seizure frequency reduction — remarkable results for a population with previously uncontrolled epilepsy.

11. Where AI Meets BCIs

AI and BCIs are becoming increasingly intertwined:

11.1 Neural Decoding with Deep Learning

The core decoding problem — translating neural spike patterns into intended movements or speech — is fundamentally a machine learning problem. Modern systems use recurrent neural networks (LSTMs), transformer architectures (BRAND, the BrainGate neural decoding framework uses a GPT-2-like architecture), and population vector algorithms. The quality of the decoder directly determines BCI performance; advances in neural decoding AI have been a major contributor to the performance improvements seen in recent years.

11.2 Foundation Models for Neural Data

In 2024, research groups at Stanford and Columbia demonstrated neural foundation models — large pre-trained models trained on multi-subject neural recording data that generalize across individuals. This is important because current BCIs require patient-specific calibration sessions (typically 30–60 minutes) before each use. A strong foundation model could reduce this to seconds or eliminate it entirely, making BCIs practical for daily use without re-calibration.

11.3 Language Models as BCI Decoders

The Francis labs at UCSF and UT Austin demonstrated that language model priors dramatically improve neural speech decoding. Rather than treating each word independently, a GPT-class language model weights decoding probabilities based on expected sentence context — the same way voice recognition systems use statistical language models to resolve acoustic ambiguity. Integration with LLM priors improved their speech BCI accuracy from ~70% to ~93% word accuracy.

12. FDA Regulation of Neural Devices

Neural implants are regulated by the FDA as Class III medical devices under the device pathway (21 CFR 882) — the highest risk class, requiring Pre-Market Approval (PMA) based on clinical trial evidence of safety and efficacy. This is a more demanding standard than the 510(k) pathway used for lower-risk devices.

Neuralink received Breakthrough Device Designation for the PRIME study's N1 implant — an FDA program that provides more frequent and intensive guidance to device developers, intended to expedite development of devices addressing serious conditions. This is not market approval; it is accelerated development support.

The pathway to full PMA approval requires: demonstrated safety over sufficient follow-up time (likely 1–2 years minimum), demonstration of clinically meaningful efficacy on predefined endpoints, evidence that benefits outweigh risks for the target population, and a manufacturing quality system inspection. Neuralink has not yet applied for PMA — the ongoing PRIME study will generate the clinical evidence base for that application.

13. Ethical, Privacy, and Security Concerns

13.1 Neural Data Privacy

Neural signals are uniquely personal — potentially encoding not just intended movements but emotional states, subconscious thoughts, and identity markers. Who owns this data? Neural data is not currently explicitly protected under HIPAA (which governs health records) or GDPR (which requires explicit consent for sensitive personal data processing). Colorado became the first US state to pass a "neural rights" law in 2024 (HB 24-1058), requiring explicit consent before neural data can be collected, stored, or sold. Advocacy groups are pushing for federal legislation.

13.2 Neural Data Security

A wirelessly connected brain implant is theoretically a cyberattack surface. "Neural hacking" — unauthorized access to a BCI to read or inject signals — is a theoretical threat that security researchers have modeled. In practice, Neuralink's Bluetooth link is encrypted, and the signal bandwidth is too limited to support complex exfiltration. However, as BCIs become higher-bandwidth and more bidirectional (including stimulation), the attack surface expands. The FDA's cybersecurity guidance for networked medical devices applies to BCIs, requiring threat modeling and regular security updates.

13.3 Enhancement vs. Therapy

Current BCIs are firmly in the restorative category — returning lost function to people with disabilities. But the technology is dual-use: the same hardware that lets a paralyzed patient type could, in principle, allow a non-disabled person to control devices more efficiently, augment memory, or accelerate cognitive performance. The ethical framework for "enhancement" BCIs — particularly if they become asymmetrically accessible based on wealth — is highly contested, invoking neurorights advocacy, disability rights considerations, and concerns about cognitive inequality.

13.4 Long-Term Biocompatibility

Implanting a foreign material in the brain triggers a chronic inflammatory response — glial scarring around electrodes that degrades signal quality over time. The optimal long-term biocompatibility of flexible polymer electrode materials (like Neuralink's threads) versus rigid silicon arrays (Utah Array) versus endovascular approaches (Synchron) remains an open scientific question. No BCI system has demonstrated stable high-channel recording beyond 5 years in humans. Long-term safety data will take years to accumulate.

14. Future Directions

14.1 Increasing Bandwidth: The "1 Million Electrode" Goal

Elon Musk has stated publicly that Neuralink's long-term target is a system with sufficiently high bandwidth to address the rate-limiting bottleneck in human-AI interaction: the ~40 bits/second information transfer rate through speech and typing. A high-bandwidth BCI could, in principle, achieve communication rates of thousands or tens of thousands of bits per second — not thought transfer but high-speed intentional signal encoding. This requires far more electrodes and a deeper understanding of how the brain encodes complex information than current technology provides.

14.2 Bidirectional BCIs: Sensory Feedback

Current BCIs are primarily one-way — reading neural signals. Bidirectional BCIs close the loop: reading motor intent and also stimulating sensory regions to provide feedback. For motor BCIs, this means the patient can "feel" something touching the robotic arm. For visual BCIs, it means stimulating visual cortex to generate perceived images. Gennaris (Monash University) has demonstrated a 9-electrode visual cortical prosthesis that generates phosphene perception in blind patients — a primitive but functional artificial "vision".

14.3 Brain-to-Brain Communication

This remains a research curiosity rather than a near-term product. Demonstrations of "brain-to-brain" interfaces (BrainNet, University of Washington, 2019) involved encoding a message in EEG signals and reconstructing it as TMS stimulation in a second subject's brain. The information transfer rate was extremely low (<1 bit/decision cycle) and relied heavily on external technology mediating the "communication". True high-bandwidth brain-to-brain communication is decades away, if physiologically feasible at all.

14.4 Non-Surgical High-Bandwidth BCIs

The holy grail of BCI research is high-bandwidth, non-invasive neural recording — eliminating the surgical risk that currently limits adoption. Emerging approaches include: optogenetics (using light-sensitive proteins in neurons to enable optical recording — currently requires genetic modification, limiting clinical translation), transcranial focused ultrasound (fUS — which can stimulate specific brain regions non-invasively), and quantum-sense magnetoencephalography (wearable MEG using optically pumped magnetometers). None of these are near clinical deployment, but each represents a plausible long-term path.

15. Frequently Asked Questions

Is Neuralink safe?

The PRIME clinical trial has not reported any serious adverse events in its four implanted participants as of early 2026. The thread retraction issue in P1 was concerning but did not cause harm. All neurosurgical procedures carry risks; the FDA's Breakthrough designation reflects a judgment that the potential benefit for people with paralysis justifies proceeding with caution. Long-term safety data (5+ years) does not yet exist.

Can Neuralink read your thoughts?

No, not in any meaningful sense. The PRIME implant decodes intended motor movements from motor cortex activity. It does not decode language, memories, emotions, or subconscious content. Decoding complex thought from neural signals — even with intracortical recording — is an unsolved research problem many orders of magnitude more difficult than the motor decoding demonstrated.

When will Neuralink be commercially available?

Neuralink has not published a commercial timeline. The PRIME study is a feasibility study; a pivotal trial for PMA approval has not been announced. Realistic estimates from neurotechnology analysts suggest 4–7 years to regulatory approval for the initial restorative indication, assuming the safety profile remains favorable.

How is Synchron different from Neuralink?

Synchron's Stentrode does not require brain surgery — it is delivered via catheter through a blood vessel, like a cardiac stent. This makes it vastly less invasive but produces lower signal quality than Neuralink's direct cortical recording. Synchron has more clinical experience (more human implants) but a potentially lower performance ceiling.

Can healthy people get a BCI today?

Not legally through any clinical trial. All current human BCI trials are restricted to people with serious neurological conditions (paralysis, ALS, locked-in syndrome). Consumer applications of intracortical BCIs for enhancement are many years away from any regulatory pathway.

16. References & Further Reading

• Willett et al. — A high-performance speech neuroprosthesis, Science (2023)
• Neuralink PRIME Study — nature.com publication (2024)
• Neuralink — PRIME Study Update (2024)
• Neurology — NeuroPace RNS 10-year outcomes
• Synchron — Stentrode Technology Overview
• OpenBCI — Open-Source BCI Hardware
• Metzger et al. — High-rate speech BCI, NEJM (2023)
• BrainGate2 Clinical Trial
• Foundation Model for Neural Activity (2024)

The most important BCI question today is not "when can I get one?" — it is "how do we ensure these technologies reach the people who need them most, with privacy protections and equity built in from the start?" The technology is moving faster than the ethics. Follow the BrainGate research blog and the Neurorights Foundation to stay current on both fronts.