Neuralink records electrical activity near selected brain cells and turns movement intentions into commands for a computer. Its current implant bypasses a damaged movement pathway; it does not repair the spinal cord.

Watch on YouTube

The YouTube player loads only when you press Play. Playing connects to YouTube, which may use cookies and storage. See Privacy.

A man plays chess with no hands

Noland Arbaugh received a Neuralink implant in January 2024. The public demonstration of him controlling a chess game followed in March. The video's opening places the chess event in January, which joins two separate events.

Arbaugh controlled the cursor without moving a mouse or speaking commands. Software translated signals from his brain into cursor movements. The chess board shows one use of that control, rather than a device that understands chess by itself.

This system is a brain-computer interface, often shortened to BCI. It connects measured brain activity to an external device. The useful result is a new way to act through a computer.

Why paralysis happens: the broken cable

Movement depends on signals passing through a network. Brain cells called neurons contribute to movement commands. Signals travel through pathways in the spinal cord and peripheral nerves before reaching muscles.

Arbaugh injured his spinal cord in a swimming accident in 2016. Damage to that pathway can prevent useful commands from reaching the limbs. Movement-related activity can remain in the brain above the injury.

The cable drawing explains this separation. It does not show every cause of paralysis. Brain injury, nerve disease, and muscle disease can also affect movement. The diagram cannot describe every case.

The implant takes information from the remaining activity and sends it along a different route. The destination is a computer, rather than the person's own hand muscles. This distinction explains both the benefit and the current limit.

How do you read a brain?

An influential cell-counting study estimated about 86 billion neurons in the adult human brain. This is an average estimate from studied brains, rather than an exact count for every person.

A neuron can produce a brief change in voltage called an action potential, or spike. Such signals commonly last roughly a millisecond. A millisecond is one thousandth of a second; the precise duration depends on the cell and measurement.

An electrode is a small conducting contact that measures electrical signals. An implanted contact records voltage changes from nearby tissue. It does not hear sound, and one contact does not necessarily correspond to one neuron.

Movement information appears across patterns of neural activity. Some neurons change their firing rates with intended movement direction. The video's left and right groups simplify a population code, in which many cells can contribute to both directions.

Recording many contacts provides a richer pattern than reading a single contact. Software can compare that pattern with attempted actions. It can then estimate which movement the person intends during a trained task.

Earlier systems established this principle before Neuralink. A BrainGate participant, Matthew Nagle, received an implant in 2004. The research report appeared in 2006 and described a 96-microelectrode recording array.

The video calls it a 100-electrode sensor. The paper's recording-array count is 96. Nagle used neural control for a cursor and other external devices, while the system connected through a skull-mounted connector.

That work showed that useful movement-related signals could remain after spinal cord injury. It did not establish that every person with paralysis would obtain the same control.

Neuralink, founded in 2016, combines implanted recording contacts, electronics, wireless communication, and a surgical robot. Its N1 implant sits within an opening in the skull, beneath closed skin. The registered study describes it as wireless and rechargeable.

The N1 design described for the PRIME study has 1,024 electrodes across 64 flexible threads. These threads are finer than a human hair. They enter brain tissue, so the contacts are not simply stuck onto the brain's outside surface.

N1 and R1 are product identifiers. N1 names the implant, while R1 names the robot that places the threads. PRIME names the early study of this system.

The thin threads are difficult to handle with ordinary surgical tools. The robot inserts them at selected locations. The operation still involves a surgical team and an opening in the skull; the robot does not make surgery risk-free.

The implant processes the recorded signals and communicates without a cable through the skin. Its battery can recharge through an external charger. Wireless charging and wireless data transfer perform different jobs.

The coin comparison gives a rough sense of the implant's size. It is not a claim that the whole system consists of a bare coin or a single electrode.

From thought to cursor: the decoder

The signal path has several steps. Each step changes the form of the information while preserving a useful relationship to attempted movement.

  1. The person attempts a trained movement or directs a cursor towards a target.
  2. Electrodes record patterns of nearby electrical activity.
  3. Implant electronics process the signals and identify useful features.
  4. A wireless link sends information to an external computer.
  5. A decoder converts the pattern into a cursor command.

A decoder is software that maps measured signals to an output. During training, the system relates neural activity to targets or attempted actions. Repeated examples help it estimate movement direction and other commands.

The video describes Bluetooth transmission and spike counts. Neuralink's published technical descriptions support that general signal-processing approach. However, software versions can change, and not every future decoder must use the identical features.

The system does not need to recover every detail of a person's private thoughts. It needs information useful for the trained task. Successful cursor control therefore does not show unrestricted mind reading.

Performance also depends on practice, signal quality, and the task. A good score in a target-selection test does not automatically predict typing speed or performance in daily life.

Neuralink describes a mouse benchmark of about 8 to 10 bits per second. In its October 2026 update listing, the company reports a participant result of 11.32 bits per second.

These are information-transfer scores from a particular testing approach. They combine aspects of target difficulty, speed, and accuracy. They are not measurements of all information produced by a brain.

The 11.32 result remains a company-reported figure in this article. The available update listing supports the claim's existence, but does not provide an independent replication. It should not become a claim that every implant user outperforms every mouse user.

What it can and cannot do

Trial participants have used neural control for computer tasks, including games and communication. Neuralink and its clinical partners also report control of an assistive robotic arm. Moving that external arm differs from restoring movement in the person's biological arm.

Computer speech needs another distinction. A person can select text through a cursor and have software read it aloud. A synthetic voice can resemble an earlier recording of that person.

That result does not itself show direct decoding of attempted speech from speech-related brain activity. Neuralink describes direct speech restoration as a separate research goal. The video's computer-voice example must not merge those approaches.

The implant does not mend the spinal cord or reconnect every damaged nerve. It creates a route from recorded brain activity to an external device. The practical value can still be substantial for someone who lacks other reliable controls.

Signals can also change over time. Neuralink reported thread retraction in its first participant and changes to its methods afterwards. Early demonstrations therefore cannot establish permanent signal quality or lifelong safety.

The US Food and Drug Administration, or FDA, regulates medical devices. Its device pathway differs from the familiar numbered phases used for many medicine trials.

An investigational device exemption, or IDE, permits a qualifying device study. An IDE is permission to collect evidence under study conditions. It is not permission to sell the device for general use.

Early feasibility studies assess initial safety and function in a small group. Later studies can provide stronger evidence for a specific intended use. Study design and regulatory requirements depend on the device and its risks.

Premarket approval, or PMA, is the FDA's approval pathway for many high-risk devices. The agency examines evidence of safety and effectiveness for the proposed use. Monitoring can continue after approval.

The video's numbered road is a simplified explanation, rather than a fixed sequence that every device follows unchanged. A larger trial also does not guarantee a successful result or approval.

The public PRIME registration describes an early feasibility study of the N1 implant and R1 robot. It lists device feasibility rather than a medicine-trial phase. Its enrollment estimate concerns that registered study, rather than every Neuralink trial worldwide.

Neuralink reported 21 participants worldwide in January 2026. That dated company total does not establish an independently verified count for October. It also differs in scope from one trial's planned enrollment.

The company reports starting its UC Davis animal-research partnership in 2017. It showed recordings from a pig in 2020 and a monkey playing a cursor game in 2021. These animal demonstrations could test parts of the system but could not establish human benefit.

The company announced permission for its first human trial in May 2023. The video also dates an earlier rejected application. This article omits that rejection date because the inspected primary records do not establish it.

The sources checked for this article describe the implant as investigational. They do not establish a general-sale approval or a date when routine hospital use will begin.

The company's near-term aim is to improve independence for people with severe movement or communication limits. Its proposed vision work, called Blindsight, would stimulate visual pathways. That is a different task from recording movement-related activity.

More recording contacts might provide more useful signals, but electrode count alone does not determine performance. Placement, long-term stability, surgery, and software also matter. A proposed future electrode target is a development plan, rather than a measured clinical result.

This article omits the video's exact future electrode target because the inspected primary records did not establish it. The broader ambition to connect people more closely with artificial intelligence remains an aspiration. Current cursor-control studies do not show that proposed future ability.

The whole picture

The important connection is brain activity, recording contacts, software, and an external action. A damaged spinal pathway can remain damaged while the person gains useful computer control. Both statements can be true at the same time.

What this means

A brain-computer interface can turn surviving movement-related signals into a practical control method. Assessing its value requires real tasks, safety follow-up, and results across participants.

Early success deserves careful description. A cursor, robotic arm, synthetic voice, and repaired spinal cord represent different outcomes. Keeping those outcomes separate makes the progress easier to understand.

FAQ

The system discussed here bypasses a damaged pathway to control external devices. It does not repair the spinal cord.

Does the implant read every thought?

No. The reported decoder uses task-related signals to estimate trained actions. Those results do not establish general mind reading.

Are the electrodes on the brain's surface?

The N1 threads enter brain tissue. Their contacts record nearby electrical activity within that tissue.

Is a trial approval the same as permission to sell it?

No. Permission to study an investigational device differs from marketing approval for a defined use.

Does the cursor speed record apply to everyone?

No. It is a company-reported result from a particular task and participant. Performance varies with the person and testing conditions.

Sources