Kenneth Shock Thought a Sentence. Neuralink Said It.

Neuralink's N1 implant converted the imagined motor signals of Kenneth Shock, an ALS patient, into audible speech on March 25, 2026.

Shock was diagnosed with ALS in January 2024. He received the 1,024-electrode N1 chip in January 2026; Neuralink trained the synthesis model on his 2020 recordings, before his voice had deteriorated, which his wife called "Original Ken."

Training ran in three stages: speaking aloud, then mouthing words silently, then imagining speech with no physical movement. The March 25 demonstration used the third stage only.

What Was and Was Not Disclosed

On March 31, Musk called the result "restoring speech." Neuralink acknowledged "delays in processing and accuracy challenges" but published no word error rate, latency figure, or vocabulary range. The absence of a number is not a stated policy; Neuralink simply has not released accuracy data.

Stanford's 2023 Nature paper, from an ALS patient with a comparable intracortical array, reached 23.8% word error rate on a 125,000-word open vocabulary at 62 words per minute. Chang's lab at UCSF hit 25% median WER at 78 words per minute the same month, using surface electrodes rather than implanted ones.

VOICE is registered with the FDA as an Early Feasibility Study, the safety tier. Breakthrough Device Designation accelerates the review clock but does not substitute for the efficacy data a Pivotal trial requires.

That 23.8% WER figure changes the math on what Neuralink must eventually file. A Pivotal application runs straight into the academic field's benchmark, and the VOICE cohort has published no accuracy number to make the comparison.

Watch the VOICE page on ClinicalTrials.gov for a Pivotal registration. EFS-to-Pivotal is the gate where accuracy benchmarks stop being optional; standard FDA timelines put it no earlier than 2027. Until then, the only published WER in the record belongs to Stanford.