Scientists at Stanford University have reported a breakthrough in neuroscience that may allow people who cannot speak to communicate more naturally. A research team has developed a brain implant capable of decoding “inner speech”, the silent words people form in their minds without speaking them aloud. The findings, published in Cell in August 2025, represent one of the first demonstrations that imagined words can be reliably translated into text through a brain–computer interface.
The study involved four participants, each of whom had lost the ability to speak due to conditions such as motor neurone disease and brainstem stroke. These individuals were fitted with tiny arrays of electrodes implanted in the motor cortex, the region of the brain that controls movement, including the movements needed for speech. The implant recorded neural activity patterns as the participants silently imagined speaking words and sentences. Artificial intelligence systems then analysed those signals to predict the intended words.
The researchers found that the brain activity produced during inner speech was noticeably weaker than when participants attempted to move their mouth or tongue. Even so, the patterns contained enough information for the system to decode them. Across a vocabulary of around 125,000 words, the system achieved an average accuracy of about 74 per cent, which the team considers a major advance given the difficulty of detecting signals that do not result in physical movement.
Previous brain–computer interface studies had focused largely on decoding attempted speech, where participants try to articulate words even though they cannot produce sound. This approach has produced higher accuracy rates, since attempted speech involves stronger signals in the motor cortex. The Stanford team’s work extends this progress by demonstrating that entirely imagined words, which leave no outward trace, can also be captured and translated. For people who may not be able even to attempt speech, inner monologue decoding could provide an alternative route to communication. The researchers placed particular emphasis on privacy and consent. To address concerns that an implant might decode thoughts without a person’s intention, the team designed a safety mechanism. Participants were asked to think of a specific “password phrase” before decoding could begin. Only after producing this phrase internally would the system activate and begin translating imagined words. In the trials, this password mechanism was highly reliable, triggering correctly more than 98 per cent of the time. The researchers stressed that the system cannot eavesdrop on thoughts continuously; it only functions when explicitly engaged by the user.
Although promising, the work is still at an early stage. The decoding accuracy, while unprecedented for inner speech, is not yet precise enough to allow fluent conversation. Sentences were sometimes reconstructed with errors, and the decoding process requires intensive computing power and calibration. The implants themselves are invasive, involving surgery to place electrodes in the brain. Long-term stability of the implants and the durability of signal quality will need to be tested further.
Nonetheless, the study offers a glimpse of what may become possible in the coming years. If refined, inner speech decoding could give people with paralysis or advanced neurodegenerative disease a way to communicate more directly and spontaneously than with current assistive technologies. The researchers imagine future systems that could produce naturalistic speech on a screen or synthetic voice output in real time, allowing conversations that resemble ordinary dialogue.
The findings also raise wider questions. Decoding silent thought touches on deeply personal aspects of human experience, and ethical safeguards will be essential. The Stanford team has emphasised that their system is designed to operate strictly on command and only for clinical purposes. They acknowledge, however, that the possibility of reading inner monologue brings new responsibilities for how such technologies are developed, regulated and used. For now, the results stand as a scientific milestone. They demonstrate that even the quietest expressions of language in the mind leave detectable traces in the brain, and that with the right tools these traces can be harnessed for communication. For people who have lost their voice, the prospect of regaining a means to express themselves more naturally may no longer belong to the realm of science fiction, but to the steadily advancing field of neuroscience.
The new study can be read here
