Here is how Neuralink works, and how brain-computer interfaces in general turn a signal from the brain into a cursor move on a screen. The short version: a sensor picks up activity from nerve cells, software learns what that activity means for one person, and the screen shows the result so the person can adjust. The harder part, which researchers keep returning to, is keeping that translation accurate over time.
Educational only, not medical advice. This article does not recommend or promote any medical product. Neuralink figures below come from company posts, not peer-reviewed papers, and are labeled as company claims.
How Neuralink works: the basic chain
Researchers describe a brain-computer interface as three parts: sensors that record brain activity, software that interprets it, and a device that acts on the commands. The device also gives feedback. The person sees on screen what the command did and can try again. Neurons, the brain’s cells, give off tiny electrical pulses, and some of that activity appears as short electrical spikes that software can count.
A spike count is just numbers, and it moves nothing by itself. Software has to learn what those numbers mean. Researchers call that software a decoder. You can think of it as a translator, and each person needs their own, because each person’s signals differ.
How Neuralink works: what studies have shown
Typing with a cursor (2017)
In a 2017 eLife study, three people with paralysis, two with ALS and one with a spinal cord injury, used the BrainGate intracortical interface to point and click on letters with a cursor. The best participant averaged about 7.8 words per minute. The authors reported typing rates 1.4 to 4.2 times those of earlier systems, and information throughput 2.2 to 4.0 times earlier performance.
Reaching and grasping with a robotic arm (2012)
In a 2012 Nature study, two people with paralysis after a brainstem stroke, in a trial called BrainGate2, used signals recorded by a 96-channel sensor in the motor cortex to control a robotic arm. One participant drank coffee from a bottle with the arm on four of six attempts. The authors noted the arm was not as fast or accurate as an able-bodied person’s arm, and that the study was small.
Both studies are peer reviewed, but they are from 2012 and 2017, and they are not Neuralink’s work. Newer research may have surpassed them.
How Neuralink works: why the translation has to be adjusted, drift
If this was shown years ago, why is repeated adjustment still needed? One reason researchers give is that brain signals change over time. In the 2012 study, for example, one participant’s signals weakened over time, with lower spike amplitudes and fewer contributing channels. When signals change, a fixed translator becomes less accurate and needs another practice session, often called recalibration.
What Neuralink says
Neuralink says twenty-one people were enrolled across its trials, as of its January 2026 update titled Two Years of Telepathy. In an October 2026 post, Pretraining on 50,000 Hours of Unlabeled Brain Data, the company says self-supervised models kept decoders working for weeks without recalibration and that it set a record of 11.32 bits per second, a grid-task score that combines speed and precision. These are company claims from a self-published update. We could not confirm them independently, and the 11.32 record depends on the company’s own comparison.
The company also reported that in its first participant, a number of the implant’s thin threads retracted from the brain in the weeks after surgery, reducing the number of working electrodes and lowering the cursor score. According to Popular Science’s report, Neuralink said later changes to its recording algorithm worked around the problem. The cause has not been confirmed.
How Neuralink works: established, claimed, and unknown
- Established by peer-reviewed studies: in 2012 and 2017, people with paralysis used implanted sensors to move a robotic arm and to type with a cursor.
- The company says: twenty-one people enrolled by January 2026, an 11.32 bits-per-second record, and decoders that keep working for weeks without recalibration.
- Could not confirm: whether the company’s results hold up independently, how long they last, and how well they work in daily life.
The honest answer: the signal is recorded, software learns to translate it, and keeping that translation accurate over time is the hard part. Safety also matters, which is why trials track unwanted medical events. For how we handle sources like these, see our editorial policy, and browse more in Technology & AI.
Frequently asked questions
How does a brain signal move a cursor?
A sensor records brain activity, software learns a person-specific pattern linking that activity to intended movement, and the cursor moves on screen so the person can see and adjust the result.
Why does the system need recalibration?
Brain signals can change over time, so a fixed translator can become less accurate and need another practice session.
Is this medical advice?
No. This is general education. Brightward is independent and is not affiliated with Neuralink or any medical organization.