A. FDA Breakthrough Device Designation and the Vision Restoration Project

On the evening of September 17, 2024, Elon Musk typed a short message at Neuralink headquarters in Fremont, California. The sentence he posted on X was only a few words long. "Blindsight has received FDA Breakthrough Device Designation."

The sentence spread through blind communities around the world in an instant. For people who had lived in darkness, this was not just regulatory news. It was a signal flare that dawn was coming.

The name Blindsight carries a paradox. In neuroscience, the term originally refers to a strange phenomenon in which patients with damage to their visual cortex cannot consciously see, yet respond to visual information unconsciously. It was no coincidence that Neuralink chose this name for its project.

What they intended to do was bypass the eye, the biological input device, entirely and beam images directly into the brain's visual cortex. Seeing without eyes. Seeing with a severed optic nerve. Seeing even if you have never known light from the day you were born. That was the promise of Blindsight.

Understanding the FDA's Breakthrough Device Designation requires some context about American medical regulation. The program was created in 2012. It is a form of priority review status granted to devices that show potential to treat conditions that are life-threatening or cause irreversible disability. With the designation, companies can work more closely with FDA specialists, gain flexibility in clinical trial design, and shorten the review timeline. But this is not market approval. It means standing at the starting line of the race, not crossing the finish line.

Getting this designation required Neuralink to accumulate years of animal testing data. Experiments were repeated in which electrodes were implanted in monkeys' visual cortices and electrical stimulation was applied to see whether the animals could perceive points of light when no actual light was present. According to the summer 2025 update, one monkey had been living healthily with the visual implant for over three years. Neuralink engineers reported that the monkey recognized and responded to artificial visual signals injected into its brain with over 66 percent accuracy.

When asked where the inspiration for this project came from, Musk always gives the same answer. Star Trek. He mentions the character Geordi La Forge from the television series. Born blind, Geordi gains a wider spectrum of vision than ordinary people through a device called a VISOR.

Musk's vision goes further. Superhuman sight capable of seeing infrared, ultraviolet, even radar wavelengths. That, of course, is a story for the distant future. As he candidly admits, early versions of Blindsight will be crude, like the low-resolution graphics of an Atari game console.

Yet even that crudeness would be a miracle for some. Around 2 billion people worldwide live with visual impairment. More than 43 million of them are completely blind. For patients whose optic nerves are damaged or who have lost their eyes, existing artificial retina technology is useless. Devices like the Argus II require some remaining retinal function to work. Blindsight leaps past this limitation. It skips the eye entirely and connects directly to the occipital lobe. A camera replaces the eye, a computer replaces the optic nerve, and the implant stimulates brain cells directly.

In early 2025, Musk mentioned a specific timeline while answering audience questions at a Wisconsin town hall event. "We will implant the visual device in the first human patient within six to twelve months." If his words hold true, the historic surgery will take place by late 2025 or early 2026. The moment a person who has never seen since birth perceives light for the first time. It would be recreating, through the power of engineering, a scene of miracle that belongs in scripture.

Skeptics have raised warnings as well. The process by which the brain interprets artificial electrical signals and recognizes them as meaningful shapes is extremely difficult. In the case of congenitally blind individuals, the visual cortex has never processed visual information. That region has most likely already been rewired to process hearing or touch. No one can guarantee whether the brain will interpret electrical stimulation as light or dismiss it as meaningless noise. The FDA designation acknowledges that this challenge is on a scientifically viable track. But the finish line is still far away.

The questions raised by the Blindsight project are not just technical ones. If this technology succeeds, we may have to rewrite the definition of seeing. Is vision only what the retina receives and the optic nerve transmits? Or if the brain processes optical information through any pathway, does that count as vision too? When a person who has never known light finally sees the world, is it the same world we see? Or is it an entirely new kind of perception? There are no answers to these questions yet. Finding them requires, first, people who open their eyes.

B. The Principles of Artificial Vision Through Visual Cortex Stimulation

In 1929, the German neuroscientist Otfrid Foerster made an astonishing discovery. While electrically stimulating a patient's occipital lobe during brain surgery, the patient reported seeing a point of light even with eyes closed. He gave the phenomenon a name: phosphene. From that moment, scientists began asking a question. If electricity can create a point of light, could enough points be produced to draw a picture?

The principle of phosphenes is straightforward. Neurons in the visual cortex evolved to receive electrical signals from the retina and interpret them as light. The critical fact is that neurons cannot distinguish the source of a signal. Whether the signal comes from the retina or from an electrode, if electrical stimulation of the right intensity and pattern reaches the neurons, the brain perceives it as light. Just as pressing a piano key causes a hammer to strike a string and produce sound, stimulating neurons with an electrode makes a point of light appear.

In the 1960s and 1970s, research into visual prosthetics using this principle began in earnest. Britain's Giles Brindley and America's William Dobelle carried out pioneering experiments. In 1968, Brindley implanted 80 electrodes in the visual cortex of a 52-year-old blind woman. The patient saw a point of light at a specific location in her visual field each time an individual electrode was stimulated. Dobelle advanced this work with a 64-channel system, enabling a patient to recognize 15-centimeter letters from a distance of 1.5 meters. It was a historic achievement that proved the possibility of visual prosthetics.

These early attempts, however, hit fundamental limits. The number of electrodes was too small. With 80 electrodes, you can produce only 80 dots. That is the equivalent of an 8-by-10 pixel display. Considering that today's smartphone screens deliver millions of pixels, you can imagine how crude the images were with the technology of that era. Surface electrodes also required large amounts of current, and prolonged stimulation risked tissue damage. The surgery itself was dangerous too: a major operation involving opening the skull, carrying significant risks of infection and bleeding.

Neuralink's Blindsight stands on this history. A half century of accumulated neuroscience knowledge has been combined with 21st-century microengineering. The biggest difference is the number of electrodes. Neuralink's N1 chip houses more than 1,024 electrodes. The S2 chip being developed for vision restoration is expected to offer more than 1,600 channels. That is 20 times the resolution of Brindley's era.

The electrode design has also changed. Flexible electrode threads, thinner than a human hair, penetrate deep into brain tissue. The visual cortex does not sit only on the brain's rear surface. Most visual processing areas are hidden inside a deep crevice called the calcarine sulcus. Neurons responsible for the center of our visual field are near the brain's surface, but those handling peripheral vision are tucked inside this crevice. Previous

surface electrodes could not reach this region. Neuralink's R1 robot can insert the electrode threads with precision, reaching deep cortical areas.

Let us trace how the system works. The patient wears special glasses with a built-in camera. The camera captures the surrounding environment. This video data is transmitted to a small computer worn at the waist. The computer extracts only the most important information from the complex imagery: edges, boundaries, the positions of obstacles. This information is converted into electrical stimulation patterns and transmitted wirelessly to the implant inside the skull. The implant stimulates neurons in the visual cortex through thousands of electrodes. The brain perceives this stimulation as points of light. Points come together to form lines, and lines come together to form shapes.

A study published in 2020 by Baylor College of Medicine presented a key breakthrough in this process. Daniel Yoshor's research team used dynamic stimulation instead of static stimulation. Rather than stimulating multiple electrodes simultaneously, they activated electrodes sequentially to trace shapes as if drawing lines. Just as you can read a letter when someone traces it on your back with a finger, the brain recognizes shapes more easily from moving stimulation than from stationary dots. Using this method, blind participants were able to identify 86 shapes per minute.

The greatest challenge, though, is not technical. It is the brain's adaptation. The brain of a person who has been blind since birth has no experience processing visual information. That region is not sitting empty; it has been repurposed for other senses. It may already be used for more precise processing of hearing or touch.

When visual stimulation is delivered to such a brain, no one knows whether the brain will interpret it as light, as sound, or as an entirely new kind of sensation.

The Neuralink team aims to solve this with artificial intelligence. They study how each patient's brain responds to different stimulation patterns and develop individually optimized stimulation strategies. It is an interactive process in which the machine adapts to the brain while the brain adapts to the machine. This is a long-term rehabilitation process that may take not weeks or months, but years. Artificial vision is not something you get by flipping a switch. It is a capability that must be acquired slowly, painfully, like learning a new language.

The principles of artificial vision ultimately converge on a single philosophical question. What does it mean to see? If a camera captures, a computer processes, an electrode stimulates, and the brain perceives, is that seeing? The answer to this question will not arrive even at the moment the first Blindsight patient opens their eyes. Only that person will know what they are experiencing, and even they will not be able to convey it to us fully. Language was built on experiences we share. A language for an entirely new kind of experience does not yet exist.

C. Convoy: Robotic Arm Control and Motor Function Restoration

One day in January 2025, a thirty-second video appeared on Neuralink's X account. On screen, a robotic arm stood before a whiteboard. It picked up a marker and began to move slowly. One stroke, two strokes, three strokes. Letters appeared. C-O-N-V-O-Y. The video offered no explanation. Just three emojis: a heart, a robotic arm, a pen. But the tech community understood immediately. Someone had moved a robotic arm and written letters using thought alone.

Convoy means to escort, to travel together. Neuralink's choice of this name for its new clinical trial carries weight. If the Telepathy project was about controlling the digital world, Convoy is about controlling the physical one. Moving beyond a cursor on a screen to picking up objects in the real world, opening doors, bringing a cup to your lips. For quadriplegic patients, this is the restoration of a dignified daily life.

The project began in November 2024. Neuralink announced it had received FDA approval for a new feasibility study. Existing PRIME Study participants would cross-enroll to evaluate their ability to control an assistive robotic arm using the N1 implant. Neuralink described this as a critical first step toward restoring not just digital freedom but physical freedom.

To understand how robotic arm control differs from mouse cursor control, you need to know the concept of degrees of freedom. A mouse cursor moves on a two-dimensional plane. It has only two directions: the X-axis and the Y-axis. A robotic arm moves through three-dimensional space. Add wrist rotation, finger movement, and grip force adjustment on top of that. The human arm has seven or more degrees of freedom. Controlling all of them simultaneously requires reading far more complex signals from the brain.

In October 2025, a participant named Nick Wray demonstrated remarkable results in the Convoy study. Paralyzed in all four limbs from ALS (Lou Gehrig's disease), he was Neuralink's eighth implant recipient. After three consecutive days of eight-hour training sessions, he began moving the robotic arm as if it were his own. He picked up a cup, brought it to his mouth, and drank water. In a dexterity test designed for stroke patients, he moved 39 cylinders onto a table in five minutes. In another test, he flipped five puzzle pieces. He described the experience as one of the most amazing things in his life.

The principles behind robotic arm control are an extension of the Telepathy project. The N1 chip records neuron firing patterns generated in the motor cortex. When a patient thinks about reaching out an arm or curling their fingers, a specific pattern of electrical signals fires. Although the spinal cord injury prevents the actual arm from moving, the brain's motor plans are still being generated. An artificial intelligence algorithm decodes these patterns in real time and relays the commands to the robotic arm.

What matters here is how the decoding algorithm learns. At first, the computer has no idea what the brain signals mean. The patient imagines a specific movement, and the neural signals produced at that moment are recorded. This process repeats hundreds, thousands

of times. The computer finds patterns in the data. The pattern for extending the arm forward, the pattern for moving right, the pattern for clenching a fist. Over time, the algorithm grows more accurate, and the patient gains increasingly natural control of the robotic arm.

Yet robotic arm control is missing a crucial element: sensory feedback. The reason we can grip a cup without crushing it is the pressure we feel in our fingertips. Controlling a robotic arm without sensation is like playing the piano with a numbed hand. Pressing the keys is possible, but controlling dynamics is not.

Neuralink is developing a bidirectional BCI to solve this problem. The idea is to attach pressure sensors to the robotic fingertips and send the information those sensors detect back to the brain's sensory cortex. When a patient picks something up with the robotic arm, the tactile sensation is converted into electrical stimulation and delivered to the brain. It is still in early stages, but if it succeeds, patients will feel the robotic arm as if it were their real one.

The ultimate goal of the Convoy project goes beyond robotic arms. It is the integration with Digital Bridge technology. The concept is to wirelessly transmit motor commands from the brain to a stimulator below the spinal cord, moving a paralyzed patient's actual limbs. A research team at the Swiss Federal Institute of Technology in Lausanne has already succeeded in making paralyzed patients walk through spinal stimulation. If Neuralink can wirelessly reconnect the severed bridge between brain and spinal cord, a patient confined to a wheelchair could stand on their own legs again.

The case of Alex, the second PRIME Study participant, is also worth noting. Paralyzed below the neck from a spinal cord injury, he received a Neuralink implant and performed everyday tasks with a robotic arm mounted on his wheelchair. He flipped switches on and off, opened doors, and picked up soft pretzels to eat. He created 3D designs in CAD software and played Counter-Strike. This is not just a step forward in medical technology. It is the process of reclaiming sovereignty over a lost life.

The questions Convoy raises run deep. Is a robotic arm part of me, or is it a tool? When a patient begins to feel a robotic arm as their own, is that arm an assistive device like a prosthetic, or an extended body? If the brain directly controls it and receives sensation back, where does the boundary lie? We may need to redefine the relationship between human and machine.

D. Breakthrough Device Designation for Speech Restoration Technology (2025)

Bradford G. Smith left behind many words before ALS took his voice. Conversations with family, phone calls, narration for YouTube videos. Those voices were preserved as digital files. In 2024, he became Neuralink's third human clinical participant. By the time the chip was implanted in his skull, he could no longer speak. Months later, he began speaking again. In his own voice.

In May 2025, Neuralink announced it had received Breakthrough Device Designation from the FDA for its speech restoration technology. The designation covers patients with severe speech impairment caused by ALS, stroke, spinal cord injury, cerebral palsy, multiple sclerosis, and similar conditions. It was Neuralink's second Breakthrough Device Designation after Blindsight. Neuralink was now building the full spectrum of neural prosthetics around three pillars: motor control (Telepathy), vision restoration (Blindsight), and speech restoration.

To understand how speech restoration works, you first need to know how speaking happens in the brain. When we talk, a region called Broca's area, the brain's language center, plans what to say. That plan is transmitted to the motor cortex, where it is converted into precise commands that move the lips, tongue, jaw, and vocal cords. At every moment, dozens of muscles coordinate with exacting precision to produce the sounds we intend. At a rate exceeding 150 words per minute.

In ALS patients, motor neurons gradually die. The nerve connections that move muscles are severed. But Broca's area and the motor cortex still function. When a patient tries to say hello, the brain still generates all the motor plans required. Those plans simply never reach the muscles. Neuralink's speech restoration technology bypasses this broken connection. It reads the speech plans directly from the brain and outputs them through an external device.

Existing augmentative communication devices relied on eye tracking or muscle signals. The system Stephen Hawking used is the most well-known example. Patients move their eyes to select letters on a screen, one at a time. This method is slow, around 10 to 30 words per minute. Natural conversation is impossible. It is communication built on waiting and patience. The other person must sit quietly until the patient finishes a single sentence. Trading jokes or having spontaneous conversation is out of the question.

Brain-computer interfaces have the potential to break through this speed barrier. Reading language signals directly from the brain eliminates the intermediate step of moving eyes or contracting muscles. Research teams at Stanford University and UC Davis have already succeeded in decoding brain signals at a rate of 62 to 78 words per minute. That approaches natural conversational speed. Neuralink aims to push this speed even higher using more than 1,024 high-density channels.

Bradford Smith's case shows how far this technology has come. Hair-thin electrode threads extend from his skull into brain tissue. 1,024 electrodes record neuron activity in real time. An AI algorithm interprets these signals. It converts the neural patterns that fire when he tries to speak into text. Other research teams have reached this point as well.

Where Neuralink went a step further is the integration with voice synthesis technology. The AI was trained on voice recordings Bradford left behind when he was healthy. His intonation, his tone, his speech habits. The AI uses this data to reproduce Bradford's voice. When text decoded from the brain is fed into the speech synthesis engine, what comes out is not a mechanical robot voice. It is Bradford's own voice.

He used this technology to narrate YouTube videos. He traded jokes with his family. In a video Musk released, he played Mario Kart with his children. Neuralink co-founder DJ Seo recalled that the moment was incredibly moving. It was the moment technology gave a person their identity back.

The FDA's Breakthrough Device Designation for speech restoration technology marks a turning point in this field. The designation is a regulatory acknowledgment that the technology has the potential to deliver clinically meaningful results. Of course, a long road remains. Decoding accuracy must improve. Error rates must come down. Long-term

stability must be verified. The technology may not work the same way for every patient. Every brain is different, and so is the progression of every disease.

But the direction is clear. Someday, people trapped in silence will speak again. In their own voices. Musk might call this not the end of language but its evolution. A world where thoughts become words the instant they form. Going further still, a stage where pure thought is transmitted to others without even the intention to move the lips. The age of conceptual telepathy. That remains a distant future.

What matters right now is the patient in the hospital bed. The person who wants to call a loved one's name but cannot. The person who wants to say thank you, I'm sorry, I love you, but has no way to say it. For them, speech restoration technology is a key that opens the prison where a soul has been locked away. The connection between brain and machine is tearing down the wall of silence and returning the resonance of lost voices to the world. The FDA's Breakthrough Device Designation is a sign of hope that this resonance can reach more people.