The patient did not try to speak. She did not move her mouth, tense her jaw, or attempt to force air through vocal cords that paralysis had already taken from her. She only imagined the words, and a wireless implant reported by the New York Times on September 14 turned that alone into fluent, real-time speech. Most brain-computer interfaces for speech work by reading the motor commands for articulation, the signals meant to move tongue, lips and larynx, and decoding those into phonemes. This system instead decodes the internal rehearsal that precedes any attempt at movement, the mental step where you compose a sentence before your body tries to say it. That distinction matters because paralysis often leaves the rehearsal signal intact even when the movement signal is degraded or absent, so a decoder built for imagined speech can work in patients whose attempted-speech signal is too weak to read.

This is the fourth brain-computer-interface speech result this desk has covered in eight days, after UC Davis's pooled-data pretraining approach and Paradromics's high-data-rate implant, and the pattern across all of them is the same: every one is a first-in-human result, tested and shown to work in exactly one patient. None has yet been repeated on a second patient with a different electrode array, which is the step that would show whether a given decoding approach generalizes or was tuned, consciously or not, to one person's cortex. For imagined-speech decoding specifically, that gap matters more than for attempted-speech systems, because there is no muscle-command signal to fall back on if the imagined-speech signal turns out to vary more than expected between patients. The company has not named a date for a second implant.