Meta introduces Brain2Qwerty v2, a noninvasive AI system that decodes brain activity into real-time text

Meta says Brain2Qwerty v2 can decode sentences from noninvasive brain recordings in real time, approaching accuracy once limited to surgical brain-computer interfaces. The system was trained on MEG recordings from nine volunteers typing about 22,000 sentences, and Meta is releasing training code while BCBL releases the v1 dataset.

The work points toward more scalable brain-computer interfaces for people who cannot communicate due to brain lesions, though MEG hardware remains expensive and impractical for everyday use. Open code and data could accelerate neuroscience research, while the use of language models highlights a broader trend of combining noisy biological signals with AI-driven context reconstruction.