Neuralink had been promising what seemed like miracles, yet its most visible results to date have focused on controlling machines with thought alone. However, something different has now occurred—something that brings the technology closer to a concept Elon Musk has long discussed: directly transforming brain activity into a spoken voice.

The man featured in this demonstration is named Terry. In 2024, he was diagnosed with bulbar-onset amyotrophic lateral sclerosis (ALS), a form of the disease that can early on affect the muscles involved in speech and swallowing. While Terry can still move his body and mouth, his ability to speak had deteriorated. According to Neuralink, the situation reached a point where even those close to him struggled to understand what he was trying to say; yet, there was one thing the disease had not erased. Inside Terry’s brain, the words remained, and that is precisely what Neuralink is attempting to access.

Terry is participating in "Voice," a clinical study investigating whether the "N1" brain implant can transform signals associated with the intent to speak into text or directly into a spoken voice. The study remains experimental, and the N1 device has not yet been approved for commercialization by the FDA or other regulatory authorities. To understand what is happening, we need to go back to the milliseconds before a person utters a word. When a person decides to speak, the regions of the brain involved in speech production generate signals that normally go on to control the muscles of the mouth, tongue, and larynx.

In certain neurological conditions, that chain is broken: the person knows what they want to say and the brain generates activity related to it, but the body can no longer transform those signals into intelligible speech. Neuralink’s concept is to create an alternative pathway; instead of relying on muscles, the implant attempts to capture neural activity directly from the brain.

The N1 uses 1,024 electrodes distributed across 64 surgically implanted, extremely fine threads to record neural activity; algorithms then seek to identify patterns in that activity and convert them into commands—or, in the case of voice, into communication. Terry’s training makes this story particularly interesting. At first, he tried to articulate words as best he could.

The algorithm observed the patterns generated as Terry attempted to speak; gradually, the system began associating specific neural patterns with the words he intended to produce. Then came the pivotal moment: Terry no longer needed to try to articulate them in the usual way. He could simply think of the words—without even moving his lips—and hear them played back. Even more importantly, the playback used his own voice, which had been cloned from recordings made when he could still speak.

Terry is not the only participant demonstrating this capability; just this week, Neuralink released another short video. A participant appears before a loved one, yet he does not utter the phrase in the conventional manner. The implant records the signals associated with the intention to speak, the system interprets them, and a voice says, "I love you." It is a brief demonstration, yet there is a vast difference between this and merely selecting letters on a screen: he is speaking—conveying identity, rhythm, and intonation—and, above all, preserving the sense that the person is truly being heard. This challenge is not unique to Neuralink.