Neuralink: How Elon Musk’s Brain Chip Works, Human Trials, Risks and What Comes Next

Neuralink Has Entered the Human Brain. What Happens Next?

A brain implant can already turn intended movement into cursor control — and, increasingly, into control of machines in the physical world. Neuralink’s real test is no longer whether this is possible, but whether it can become safe, durable medicine rather than an extraordinary experiment.

Neuralink is developing an implanted brain-computer interface designed to translate neural activity into commands for computers and assistive devices.

Imagine being unable to move your arms or legs while the intention to move them remains perfectly clear. The command still forms in the brain — reach for the cup, move the mouse, type the sentence — but somewhere between cortex and muscle, the route has been broken.

Neuralink is trying to build another route.

That sounds like science fiction until you watch Noland Arbaugh, Neuralink’s first human participant, move a cursor, play chess and navigate a computer without touching a mouse. His implant, placed in January 2024, turned patterns of neural activity into digital movement. The remarkable part is that this worked. The less glamorous — and more important — part is that neuroscience had already proved the basic principle years earlier. Implanted electrodes had controlled computer cursors by 2006 and robotic arms by 2012.

Neuralink’s wager is therefore not that the brain can control a machine. We already know it can. The wager is that an experimental brain-computer interface can be redesigned into something closer to a medical platform: fully implanted, wireless, rechargeable, usable at home, and inserted with a robot rather than assembled around a patient in a laboratory.

By 2026, that distinction is what makes the company interesting. Neuralink is operating human studies in several countries, extending its work from computer control toward robotic assistance and speech, and preparing a visual-prosthesis program called Blindsight. Yet the evidence is still young. The participant numbers are small, the trials are early-stage, and many of the most striking updates come from Neuralink itself rather than completed peer-reviewed clinical reports.

So the useful question is no longer the one that dominates headlines: can a brain control a computer? It can. The harder question is whether Neuralink can make that connection stable enough, safe enough and useful enough to survive the transition from demonstration to medicine — and what follows if it does.

Neuralink did not invent the brain-computer interface — and that matters

Brain-computer interface, or BCI, sounds like a phrase invented for a cyberpunk novel. In practice, it describes something more specific: a system that detects useful patterns in neural activity and converts them into commands for an external device. The field is much older than Neuralink, and that history matters because it gives us a baseline. Without it, almost every Neuralink demo looks unprecedented. With it, we can see more clearly where the company is genuinely pushing the field forward.

A BCI does not need to understand a person’s mind in any general sense. It needs a repeatable signal. If groups of neurons change their firing in a reliable way when someone intends to move a hand to the right, software can learn that pattern and map it to a cursor moving right. That is already extraordinary — but it is very different from decoding a hidden stream of thoughts.

The lineage is well documented. In 2006, the BrainGate team reported in Nature that a man with tetraplegia could use signals from his motor cortex to control a computer cursor and simple devices. In 2012, people with profound paralysis used intracortical signals to reach and grasp with a robotic arm. By 2023, research groups were decoding attempted speech into text at useful speeds, while one system also generated synthetic speech and facial-avatar movements. Neuralink arrived after these breakthroughs, not before them.

It also arrived into a field with competing ideas about how much invasiveness is worth accepting for better signals. Synchron reaches the brain through a blood vessel rather than penetrating cortex directly. In its first-in-human SWITCH study, four participants used the system for hands-free computer tasks without serious device-related adverse events during the 12-month study period. Precision Neuroscience is developing ultra-thin surface arrays. Paradromics is pursuing a high-data-rate penetrating interface, with communication as a major goal.

That makes the Neuralink strategy easier to define. It is betting that a high-channel-count implant placed inside the cortex can deliver richer signals — and that robotics, miniaturization and wireless engineering can make an invasive approach practical enough for daily life. The scientific question is no longer whether penetrating electrodes can work. It is whether they can work for years in real people.

There is no single way to connect a computer to the brain. Neuralink uses flexible penetrating threads, while other BCI systems rely on rigid intracortical arrays, electrodes inside blood vessels or arrays placed on the cortical surface.

What is actually inside a Neuralink implant?

The N1 Implant is designed to sit flush in a circular opening in the skull, hidden beneath the scalp. From it extend 64 extremely thin, flexible polymer threads carrying a total of 1,024 electrodes. Those electrodes sit close to neurons in the cortex and record tiny changes in electrical activity — signals measured in microvolts, far below anything a person can consciously feel.

The flexibility is not cosmetic. The brain is soft tissue, and it is never perfectly still: it shifts subtly with breathing, blood flow and movement. Older rigid arrays create a mechanical mismatch between hard electronics and soft biology. Neuralink’s threads are intended to move more naturally with the tissue. The price of making them hair-thin and flexible is that they are too delicate to insert reliably by hand.

That is why the implant is only half of the Neuralink system. The other half is the R1 surgical robot, which uses imaging and precision positioning to place individual threads while trying to avoid surface blood vessels. Neuralink’s 2019 technical paper described an earlier platform with thousands of electrodes and automated insertion. The human N1 uses 1,024 electrodes, but the underlying philosophy is the same: make the neural interface dense, make the electronics compact, and make implantation repeatable enough that it can eventually move beyond a one-off laboratory procedure.

Once implanted, the device amplifies and digitizes the neural signals and sends them wirelessly to external software. Its internal battery is charged inductively through the skin. There is no permanent cable protruding from the head during normal use — an unglamorous detail, but one that matters enormously if a BCI is ever supposed to become part of ordinary life.

The N1 implant records electrical activity through flexible electrodes placed in the cortex. Software then decodes those neural signals and converts intended movements into digital commands.

How can an intention move a cursor?

The easiest way to misunderstand Neuralink is to imagine that the implant somehow listens to a complete thought and translates it. What it actually does is narrower and, in some ways, more interesting: it looks for neural patterns connected to intended action.

Before a hand moves, populations of neurons in motor-related areas change their activity. A spinal cord injury can sever the route from brain to muscle without erasing the intention itself. The command is still being generated; it simply cannot reach its usual destination. An implanted BCI records a small part of that activity and sends the useful information along a different path.

That path has to be learned. During calibration, a participant imagines or attempts movements while the system observes the accompanying neural patterns. Machine-learning models then begin mapping those patterns to commands: move the pointer, click, select a letter, operate an assistive device. The better the mapping becomes, the more direct the control can feel.

There is a subtle point here that demos often hide: the machine is learning the user, but the user is learning the machine too. People discover which imagined movements produce the most reliable response; decoders adapt to changing signals; habits form on both sides of the interface. A working BCI is therefore not simply a reader. It is a new skill shared between a nervous system and an algorithm.

That is also why the phrase “mind reading” is usually more confusing than helpful. Neuralink is not eavesdropping on an unrestricted inner monologue. It is decoding selected neural activity for trained tasks. Even speech BCIs depend on targeted brain regions, deliberate attempts to speak and models trained to interpret those signals. The distinction may sound technical, but it is the difference between a remarkable neuroprosthesis and a fictional telepathic device.

For the AI side of this process — neural networks, machine learning and the difference between biological and artificial “neurons” — see Next Horizon’s Artificial Intelligence Explained: From Neural Networks to AI Agents.

The first human implant revealed both the promise and the problem

When Noland Arbaugh received the first Neuralink implant in January 2024, the story immediately acquired a human scale. Arbaugh had been living with quadriplegia after a diving accident. With the implant, he learned to move a cursor through neural activity and later used it for gaming, browsing and other everyday computer tasks. For the first time, Neuralink was no longer demonstrating its system on a stage or in an animal study. Someone was building it into his daily routine.

Then the implant developed a problem — and the experiment became more informative.

Within weeks, a number of the electrode threads retracted from the brain, reducing the amount of useful neural data. Arbaugh later said that roughly 85 percent of the threads had moved out of position. Neuralink responded in software, changing its signal processing and decoder so that the remaining channels could carry more of the workload. His performance recovered enough for continued use.

It is tempting to tell that episode as either failure or triumph. Neither version is very useful. The first participant exposed a mechanical weakness, exactly the kind of problem early-feasibility studies are meant to uncover. At the same time, the recovery showed something unusually powerful about modern neurotechnology: hardware performance is not the whole system. Adaptive software can sometimes rescue useful function from imperfect biological signals. But it cannot make long-term mechanical stability optional. A device intended to live in the brain for years must stay where it was placed.

Neuralink altered the surgical approach for later participants, including steps intended to reduce brain movement and decrease the gap between implant and cortical surface. The company reported no thread retraction in its second participant. In a January 2026 update, it said later procedures had generally produced better signal quality and described additional mechanical changes under study. Those are encouraging engineering responses, but they still need time — and larger numbers of participants — to become evidence of durability.

The same 2026 update pointed toward the next hardware generation: roughly 3,000 electrodes instead of about 1,000, and research into inserting threads through the dura, the protective membrane around the brain, with the aim of making surgery simpler. More electrodes could increase the amount of usable information. A simpler procedure could expand access. Neither improvement is proven merely because it appears on a roadmap.

Where Neuralink actually stands in 2026

Program	Goal	Evidence/status
PRIME / Telepathy	Computer and device control for people with severe paralysis	Active early-feasibility human study; U.S. registry listed 15 participants in the study design.
CONVOY	Control assistive devices such as a robotic arm	Early-feasibility study; Neuralink has shown participants using an assistive robotic arm.
CAN-PRIME / GB-PRIME / UAE-PRIME	Expand safety and functionality studies internationally	Trials launched in Canada, Great Britain and the UAE; Canada performed the first Neuralink surgeries outside the U.S. in 2025.
VOICE	Decode intended speech for people with severe speech impairment	Active program; Neuralink says it is targeting conversational-speed communication. Independent BCI research has already demonstrated high-speed speech decoding in small studies.
Blindsight	Create visual perception by stimulating visual cortex	FDA Breakthrough Device designation; listed by Neuralink as an upcoming trial, not an approved consumer treatment.

The most important words in Neuralink’s current trial descriptions are easy to overlook: early feasibility. These studies are asking whether implantation can be performed with acceptable safety, whether the device can keep functioning, and whether participants can gain meaningful control. They are not the kind of large clinical trials used to establish a mature treatment for a broad population.

What has changed is scale and geography. Neuralink expanded from the United States into Canada, Great Britain and the United Arab Emirates. Surgeons at University Health Network performed the first Neuralink implantations outside the U.S. in August and September 2025. In Britain, UCL and UCLH reported that the first GB-PRIME participant began cursor control the day after surgery. Those milestones show that the procedure is becoming reproducible across centers rather than remaining a single-site demonstration.

But reproducible is not the same as routine. Neuralink is still at the beginning of the clinical story.

A cursor is only the beginning

It is easy to underrate cursor control because most of us perform it unconsciously. For someone who cannot reliably use their hands, however, a cursor is access: to conversation, work, entertainment, banking, education and the systems that increasingly organize everyday life. Digital independence is not a flashy demo outcome. It can change how much of a day belongs to the person rather than to a caregiver.

The limitation is obvious the moment the screen is turned off. A cursor cannot lift a cup.

That is where CONVOY, Neuralink’s study of assistive-device control, becomes important. The company has shown participants operating robotic systems for tasks such as drawing and feeding. Conceptually, the step sounds simple: instead of converting intended movement into two-dimensional cursor coordinates, convert it into commands for a machine in the physical world.

In practice, the difficulty increases dramatically. A cursor has a handful of degrees of freedom. A human arm coordinates shoulder, elbow, wrist and fingers while the brain simultaneously uses vision, touch and a continuous sense of where the limb is in space. A robotic arm controlled from the brain has to be accurate enough to be useful, fast enough not to become exhausting, and safe enough to move near a face, food, furniture and other people.

That is why the deeper destination is probably bidirectional. The system must not only read an intention to move; it may eventually need to send information back as artificial touch, pressure or position. Research outside Neuralink has already shown that electrical stimulation of the nervous system and cortex can produce useful sensations. The most capable future neuroprosthesis may be less like a remote control and more like a new sensory-motor loop.

Brain-computer interfaces could eventually move beyond cursor control, allowing people with paralysis to operate robotic arms and perform everyday physical tasks through neural signals.

VOICE could matter more than cursor control

For a person with advanced ALS or a severe brainstem injury, the most devastating loss may not be the inability to operate a mouse. It may be the shrinking of communication itself — first slower speech, then a few remaining movements, and eventually perhaps no reliable way to express a sentence at all.

This is where the wider BCI field has made some of its most striking progress. In 2023, a Stanford-led team reported a speech neuroprosthesis that decoded attempted speech at 62 words per minute using intracortical arrays. A UCSF-led group reported a surface-electrode system with a median text-decoding rate of 78 words per minute and demonstrated synthetic speech and facial-avatar control. The studies were small, but they showed that the machinery for speech can remain decodable years after the body loses the ability to produce it normally.

Neuralink entered that race with a speech-restoration program that received FDA Breakthrough Device designation in 2025 and was later described as the VOICE trial. In its January 2026 update, the company said it was aiming for communication around 140 words per minute — much closer to normal conversation — and showed a participant typing by imagining finger movements.

The number needs to be read correctly: 140 words per minute is a target, not a demonstrated Neuralink clinical result. Still, it reveals where the company thinks the technology can go. A BCI that restores fast, flexible communication could be more consequential than one optimized mainly for pointer control. For someone approaching a locked-in state, generating their own words in real time is not a productivity feature. It is a form of autonomy.

Accuracy, however, is where the philosophical problem becomes an engineering one. A decoder can be impressive in a paper and still be frustrating in a conversation. Language models can correct probable errors and predict likely words, but the more aggressively they predict, the more carefully the system must preserve authorship. At some point the question stops being only, “Did the computer understand the signal?” and becomes, “Whose sentence is this — the user’s, or the model’s best guess?”

Blindsight asks a harder question: can a machine write into the brain?

Blindsight is the most dramatic of Neuralink’s near-term programs because it reverses the direction of the interface. Telepathy mainly records neural activity. A visual prosthesis would have to stimulate the brain in precisely controlled patterns and create perception from the outside in.

The basic idea is grounded in real neuroscience. The visual cortex contains organized maps of visual space, and electrical stimulation can produce phosphenes — perceived points or flashes of light — even without normal input from the eyes. In principle, a camera could capture the scene, software could convert it into a stimulation pattern, and an implant could deliver that pattern directly to visual cortex.

What that would feel like is where popular imagination runs ahead of evidence. “Restoring sight” does not necessarily mean reconstructing the rich, continuous image experienced by a healthy visual system. The first useful artificial vision could be sparse and unfamiliar: points, edges, contrasts and motion cues that a person gradually learns to interpret for navigation or object detection.

Neuralink announced in 2024 that Blindsight had received FDA Breakthrough Device designation. That status can accelerate interaction with the agency for technologies addressing serious conditions; it is not FDA approval and it is not proof of efficacy. On Neuralink’s current trials page, the vision study remains an upcoming program. For now, Blindsight is scientifically plausible, clinically unproven and potentially transformative — all three at once.

The real risks are less cinematic than mind control

The popular fears around brain implants tend to be futuristic. The immediate risks are far more ordinary — surgery, biology, hardware and dependency — and that makes them no less serious.

Implantation is neurosurgery. Opening the skull and placing material into or near brain tissue carries risks including bleeding, infection, seizures, tissue injury, anesthesia complications and postoperative problems. A surgical robot can improve precision and consistency. It cannot make an invasive brain procedure risk-free.

Then there is the fact that living tissue reacts. The brain does not treat an electrode as a neutral object simply because engineers designed it carefully. Immune responses and scar tissue can alter the interface and reduce signal quality over time. Flexible threads are intended to reduce mechanical stress, but chronic stability is something that must be demonstrated across years, not inferred from a few good months.

Hardware introduces another clock. Threads can move. Electronics can fail. Batteries age. Future upgrades may require another operation. In consumer technology, replacing obsolete hardware is irritating. In neurotechnology, an upgrade can mean reopening the skull. That changes the meaning of product cycles entirely.

Software is part of the medical device too. Neural decoders can improve through updates — the first participant’s recovery after thread retraction is a vivid example — but that also means useful function depends on a chain of hardware, firmware, models, wireless links and external computers. Every additional layer can improve performance, and every layer can fail.

The least visible risk may be dependence on the company itself. A person with an implanted interface could rely on its manufacturer for software, replacement parts, clinical follow-up, security patches and perhaps future surgical support. The ethical obligation does not end when a study participant leaves the operating room; in some ways, that is when it begins.

Neural privacy may become a new kind of privacy

Today’s Neuralink is not extracting a secret diary from someone’s cortex. But neural data is unusual because it sits closer to the processes that generate action than most information we have ever collected at scale.

Traditional personal data usually records what happened after a choice: the page we opened, the message we sent, the place we visited. Neural signals can capture biological activity that precedes an action. As decoders become more capable, the line between an intentional command and information inferred from the surrounding neural activity may become increasingly important.

That creates questions that do not fit neatly into existing privacy law. Who owns the raw recordings? How long can they be stored? Can they be reused to train future models? Could an insurer, employer or court ever seek access? And what happens if data collected today reveals more in five years because decoding algorithms have improved? Consent is usually given for what data means now. Neural data may change meaning after it has already been collected.

Cybersecurity is the less philosophical version of the same problem. A wireless medical implant that depends on software updates must be designed against unauthorized access. The realistic threat is not a movie villain remotely taking possession of someone’s mind. It is corrupted data, disrupted device function, malicious control of connected equipment or exposure of information that a user reasonably expected to remain private.

A 2026 Nature Neuroscience commentary on implantable BCIs argues for safeguards that sound almost mundane: meaningful clinical purpose, honest separation of research from treatment, and long-term support for participants. Those principles become more important, not less, if BCIs move from a few highly monitored volunteers into ordinary medicine.

The animal-testing controversy deserves precision, not slogans

Neuralink’s preclinical program has faced sustained scrutiny over animal welfare and research practices. Reuters reported concerns from former employees and regulatory records about the pace and conduct of some experiments. In a 2023 inspection of Neuralink’s California animal facility, the FDA later identified record-keeping and quality-control deficiencies, including missing calibration records and documentation problems. The agency did not report comparable findings at the Texas site and said the available information was sufficient to support the company’s human trial.

The distinction is important. A regulatory deficiency should not be waved away, but it is also not evidence that every Neuralink result is invalid. Likewise, permission to begin a human study does not mean a device has been proven safe for mass use. The sensible position is less dramatic and more demanding: keep following the evidence as the participant numbers, follow-up periods and independent publications grow.

Neuralink’s biggest advantage may not be the implant

The most distinctive thing about Neuralink may be that it is not really building one device. It is building an entire stack.

The company controls the electrode threads, implant electronics, wireless link, charger, surgical robot, signal-processing pipeline, user software and much of the clinical workflow. That makes iteration possible across layers. When the first participant lost useful channels, software changes improved control. When thread retraction emerged, the surgical approach changed. If future versions need more channels, the implant, robot and decoding software can evolve together rather than as separate products.

That vertical integration resembles strategies used in other complex technologies, but the comparison has limits. An implanted BCI cannot be iterated like a smartphone app. Every hardware revision touches surgery, tissue and long-term clinical responsibility. A human brain is not a beta environment.

This may be Neuralink’s central cultural challenge: combining Silicon Valley’s appetite for rapid iteration with medicine’s demand for evidence, traceability and patience. The first culture asks how fast something can improve. The second asks how confidently we know it is safe. A successful brain interface will need both answers at the same time.

What Neuralink still cannot do

It cannot read arbitrary thoughts as complete sentences. It cannot upload memories, download knowledge or make a person generally more intelligent. It does not merge a user with an AI model. Those ideas are not simply unfinished features waiting for the next software update; they are separate scientific problems, some of which we do not yet know how to formulate precisely.

Scale helps explain why. The human brain contains roughly 86 billion neurons. Even a future Neuralink with 3,000 electrodes would sample only a minuscule, highly local fraction of that activity. Increasing channel count can improve bandwidth and control, but collecting more signals is not the same as understanding the neural code behind memory, identity or abstract thought.

Consciousness makes the gap even larger. Recording activity associated with movement or attempted speech does not mean an implant has located the self, a memory, an emotion or subjective experience. Neuroscience still lacks a complete account of how conscious experience emerges from brain activity. A BCI can become clinically powerful long before it becomes philosophically omniscient.

For the deeper version of that question, see Next Horizon’s Can AI Become Conscious? What Science Actually Says in 2026.

Musk’s long-term vision is much bigger — and much less certain

Neuralink is explicit about wanting to go beyond one medical indication. The company talks about a generalized brain interface and, eventually, a “whole-brain interface” that could both read from and write to many parts of the brain. Elon Musk has repeatedly connected that ambition to a future symbiosis between humans and advanced AI.

The argument begins with bandwidth. Humans consume information quickly through vision and hearing, but our output into computers is comparatively slow: fingers on keyboards, thumbs on screens, speech at human speed. A direct neural interface could, in theory, reduce the friction between forming an intention and operating a digital system.

But the brain is not a USB port waiting for someone to increase the transfer rate. Motor intention is comparatively tractable because researchers know useful places to record and can measure success directly: did the cursor move where the person intended? Memory, abstract reasoning and general cognition are distributed, dynamic processes that we only partly understand. Writing useful information back into those systems is harder still.

There is another conceptual trap in the phrase “merge with AI.” Even a perfect high-bandwidth connection to an AI system would not automatically transfer the model’s internal capabilities into the biological brain. A smartphone gives us near-instant access to the internet without turning its owner into a server farm. A neural interface could make access faster and more intimate without erasing the distinction between human cognition and machine computation.

That does not make the vision meaningless. It makes it distant. Between cursor control and cognitive augmentation lie unsolved problems in neuroscience, surgery, materials science, computing, security, ethics and regulation. Neuralink’s medical work may eventually become the foundation for something much broader — but it should not be confused with that future today.

Today’s implanted BCIs are primarily medical tools for restoring communication and control. Direct high-bandwidth interaction between the human brain and advanced AI remains a much more speculative possibility.

The scientific test Neuralink still has to pass

Neuralink has already succeeded at something outside the clinic: it made implanted brain-computer interfaces visible to millions of people who had never heard of BrainGate, ECoG arrays or neural decoders. Public attention can attract talent and capital to a field. It can also make early results feel more mature than they are.

The tests that matter now are slower and less cinematic.

Does the implant remain useful year after year? Does signal quality stay stable? How often must a user recalibrate? How much additional independence does it provide compared with eye tracking, sip-and-puff controls, voice systems and other assistive technologies? What complications appear when the participant pool grows from dozens to hundreds? Can hospitals outside a few elite centers perform the procedure safely? And eventually: what does all of this cost, and who pays for surgery, rehabilitation, upgrades and lifetime support?

None of those questions makes a viral headline. Together, they determine whether Neuralink becomes a medical platform or remains an extraordinary experimental technology.

History gives reason for caution. Implantable BCIs have been tested in humans since the late 1990s, yet a 2025 review in Nature Reviews Bioengineering noted that no implantable BCI had reached ordinary medical-device market approval for restoring communication or movement. That is a useful reminder: a technology can work scientifically for years and still struggle to cross the final distance into routine care.

Neuralink is attacking that translation problem more directly than many academic systems did: smaller hardware, wireless use, robotic surgery, manufacturing and a user experience designed for life outside the lab. That is its opportunity. The danger is that engineering progress and public expectation can move faster than clinical evidence.

What happens next is likely to be less futuristic — and more useful

The most realistic near-term future for Neuralink is medical, not transhumanist. That may sound less spectacular than the promise of human-AI fusion, but it is also where the technology could matter first.

If current trials continue to show acceptable safety, the next gains are likely to be practical: better computer control for people with paralysis, faster communication for people losing speech, and richer control of assistive devices. Speech decoding could be especially important for ALS and brainstem injuries. Robotic systems could convert intention into physical independence. Visual-cortex stimulation may eventually provide useful, if limited, perception for some people with profound blindness.

The hardware will keep changing alongside those applications. Neuralink is already discussing higher electrode counts, improved thread retention and surgical techniques intended to reduce complexity. Competitors are making different bets — lower invasiveness, surface recording, vascular access, greater bandwidth or narrower clinical specialization. There may never be one universal BCI architecture, because the safest way to restore speech may not be the best way to control a robotic limb or create artificial vision.

The larger frontier begins only after those medical problems are solved. Moving from decoding a few useful variables to interacting with memory, reasoning or complex internal states would require not just better engineering, but a much deeper understanding of the brain itself. Neuralink can accelerate that journey. It cannot skip it.

Conclusion: the real Neuralink story is smaller than mind reading — and more profound

Neuralink is usually introduced with the vocabulary of science fiction: telepathy, mind reading, human-AI fusion. The language makes the future sound enormous, but it can have the strange effect of making the present achievement look small.

The present achievement is not small. A person can intend to move a hand that no longer moves, and a machine can detect enough of that intention to move something else in the world. A cursor crosses a screen. A robotic arm responds. An attempted word becomes text. A signal that once ended at an injured spinal cord finds another exit.

We have not decoded the human mind. We have begun learning how to reroute tiny pieces of it.

That distinction is the reason Neuralink is worth watching. Its first real revolution may have nothing to do with turning healthy people into superhumans. It may be giving people who have lost a channel to the world another one — first to a screen, then perhaps to a voice, a robotic limb or a visual field.

And if those channels eventually become stable, bidirectional and high-bandwidth, the philosophical questions will arrive without needing to be exaggerated. When intention can travel directly from neurons into machines — and information can travel back — where does the body end, where does the tool begin, and how much of that boundary are we actually prepared to redesign?

FAQ

Is Neuralink already available to the public?

No. Neuralink’s implants are investigational medical devices used in clinical studies. They are not consumer products and have not received broad marketing approval for routine use.

Can Neuralink read thoughts?

Not in the everyday meaning of the phrase. Current systems decode trained patterns such as intended movement and, in research programs, attempted speech. They do not continuously translate all private thoughts into language.

How many electrodes are in the current N1 implant?

Neuralink describes the current N1 as using 1,024 electrodes distributed across 64 flexible threads. The company has discussed future versions with around 3,000 electrodes.

What is Telepathy?

Telepathy is Neuralink’s name for its BCI program intended to let people with severe paralysis control computers, phones and eventually physical assistive devices using neural activity.

What is Blindsight?

Blindsight is Neuralink’s proposed visual prosthesis program. The concept is to stimulate visual cortex directly to create useful visual perception for people with severe vision loss. It has FDA Breakthrough Device designation, but it remains investigational.

Did Elon Musk invent brain-computer interfaces?

No. Human implantable BCI research predates Neuralink by decades. Neuralink’s contribution is an integrated approach combining flexible electrode threads, a fully implanted wireless device, a surgical robot and consumer-style software engineering.

What is the biggest obstacle for Neuralink?

Long-term clinical reliability. A useful implant must remain safe, stable and functional for years while surviving tissue responses, mechanical movement, hardware aging and software changes.