A new brain–computer interface (BCI) can decode people’s silent thoughts — but only if they choose to unlock it with a preset password. The breakthrough, reported in Nature and published in Cell on 14 August, could make mind-reading devices safer and more practical for people who struggle to speak.
The system translates internal speech — the inner dialogue in a person’s mind — into text in near real time, achieving up to 74% accuracy in tests with volunteers who have difficulty speaking due to stroke or motor neuron disease. Crucially, it begins decoding only after the user silently thinks of a chosen keyword, such as “Chitty-Chitty-Bang-Bang.” This password mechanism ensures that private thoughts remain private.
“Privacy is a critical issue for speech BCIs, and this password approach provides a simple but powerful safeguard,” said Sarah Wandelt, a neural engineer at the Feinstein Institutes for Medical Research, who was not involved in the work.
BCIs have emerged as a promising way to restore communication in people with paralysis or speech impairments. Most systems to date rely on users attempting to speak aloud, but that can be exhausting. Internal-speech BCIs — which decode imagined speech directly from brain signals — offer an alternative, though they risk picking up words that users never intended to express.
To tackle this, lead researcher Erin Kunz and her team at Stanford University recorded brain signals from electrodes implanted in the motor cortex of four participants. Using artificial intelligence, the researchers trained the system to recognize the phonemes — the smallest units of speech — embedded in these signals. Language models then assembled the phonemes into words and sentences from a vocabulary of 125,000 entries.
In tests, the device correctly interpreted both imagined sentences and even spontaneous self-talk, such as counting silently. When the password feature was activated, the system recognized it with more than 98% accuracy before beginning to decode.
“This study is technically impressive and also meaningful,” said Silvia Marchesotti, a neuroengineer at the University of Geneva. “It highlights the differences between attempted and internal speech and shows a way forward for safer, real-world BCIs.”
The team now plans to improve the device’s accuracy and explore brain regions beyond the motor cortex to capture a wider range of speech signals. If successful, the technology could open new doors for people with severe communication difficulties — while keeping their innermost thoughts secure.
