Meta's new AI can turn your thoughts into text without a brain implant

Imagine composing a message without lifting a finger, moving your lips or even making a sound. Instead, your thoughts alone are transformed into written words on a screen.

It sounds like science fiction, but Meta believes it has taken a meaningful step in that direction. The company has introduced Brain2Qwerty v2, a new artificial intelligence system capable of translating brain activity into text without requiring electrodes to be implanted inside the brain.

The research is still confined to laboratories and is far from becoming a consumer product. Yet it represents one of the strongest demonstrations so far that non-invasive brain-computer interfaces could eventually help people who lose the ability to communicate after a stroke, traumatic injury or neurological disease.

What is Brain2Qwerty v2?

Brain2Qwerty v2 is Meta's latest attempt to build a brain-to-text system that works without surgery.

More from Tech

Unlike invasive brain-computer interfaces, which rely on electrodes implanted directly into the brain, Meta's approach uses magnetoencephalography (MEG). The technology records tiny magnetic fields generated by brain activity using sensors placed outside the head, eliminating the need for any surgical procedure.

Meta says Brain2Qwerty v2 is its most advanced end-to-end decoding system yet, capable of reconstructing complete sentences from brain recordings in near real time.

To train the model, researchers recruited nine volunteers. Each participant spent around 10 hours inside an MEG scanner while typing approximately 22,000 sentences. Those recordings enabled the AI to learn how patterns of brain activity correspond to words and sentences.

How AI helped improve accuracy

One of the biggest differences between Brain2Qwerty v2 and earlier systems is the way it processes information.

Previous research often depended on manually identifying specific neural signals before attempting to decode language. Meta instead built an end-to-end deep learning system that learns directly from raw brain activity.

The company also adapted large language models to work with neural data. By understanding the context and meaning of language, the AI can make better predictions even when brain signals are incomplete or noisy.

Meta said AI agents were also used during development to explore different optimisation strategies for the decoding pipeline before researchers selected the best-performing training setup.

The results represent a significant improvement over previous non-invasive systems.

According to Meta, Brain2Qwerty v2 achieved an average word accuracy of 61 per cent across participants. Earlier non-invasive approaches typically reported accuracy levels of around 8 per cent.

Performance was even stronger for some volunteers. The best-performing participant reached 78 per cent word accuracy, while more than half of the decoded sentences contained no more than a single incorrect word.

Researchers also found that accuracy improved as the model was trained on larger datasets, suggesting further gains may be possible.

Why this research matters

Despite the progress, Brain2Qwerty v2 is not about replacing keyboards or smartphones.

Meta says its primary objective is to help people who can no longer communicate through speech because of medical conditions such as stroke or neurological disorders.

Today, the most accurate brain-computer interfaces rely on implanted electrodes. While those systems have already enabled some patients to communicate using AI-powered neuroprostheses, they require complex brain surgery and remain suitable for only a limited number of people.

"Our noninvasive approach can help bridge that gap," Meta said while announcing the research.

To encourage further work in the field, the company is releasing the complete training code for both Brain2Qwerty v1 and v2. Its research partner, the Basque Center on Cognition, Brain, and Language (BCBL), will also make the original Brain2Qwerty v1 dataset publicly available.

Meta hopes the project will contribute to broader efforts to build open brain models and accelerate research into diagnosing neurological conditions and restoring communication for people who have lost their voice.