Meta’s Brain2Qwerty v2 Translates Brain Activity Into Text, No Surgery Required

6 min read · Last updated 2026-06-30

Meta has shown an AI system that reads what a person is trying to type directly from their brain activity, and it does so without any surgery. The company introduced Brain2Qwerty v2 this week, a non-invasive system that records neural signals with a helmet-like scanner and reconstructs the sentences a person intends to write. Meta says the research is aimed at helping people who have lost the ability to communicate because of brain lesions.

Why It Matters

Most high-performing brain-computer interfaces still rely on electrodes implanted directly in the brain, which means surgery, and surgery is hard to scale. It carries real risk, and implants are difficult to maintain over time. A non-invasive system that approaches the accuracy of implanted devices would remove the single biggest barrier to giving people who cannot speak or type a way to communicate. Meta says Brain2Qwerty v2 approaches levels of accuracy previously reached only with surgical techniques, which is why the announcement matters well beyond the lab.

What’s New and How It Works

The system records brain activity with a magnetoencephalography, or MEG, scanner, a non-invasive imaging device that looks like a helmet and is common in neuroscience research. Those raw neural signals feed into an end-to-end deep learning model that reconstructs the sentences the person is trying to type. Meta improves accuracy further by fine-tuning large language models on neural data, letting the system use semantic context to make sense of noisy recordings.

“Instead of relying on hand-crafted pipelines to detect neural events, we use end-to-end deep learning to decode directly from raw brain signals,” Meta wrote. The company trained the model on about 22,000 sentences from nine volunteers, each recorded for 10 hours while wearing the MEG device and actively typing.

The Numbers

• 61% average word accuracy, up from roughly 8% for previous non-invasive methods.
• About 22,000 training sentences, from nine volunteer participants.
• 10 hours of MEG recording per participant, captured while actively typing.
• A $5 million fund for open neuroscience datasets, part of Meta’s Digital Brain Project.
• Code for v1 and v2 released, with the v1 dataset released by Meta’s research partner.

“Instead of relying on hand-crafted pipelines to detect neural events, we use end-to-end deep learning to decode directly from raw brain signals.”

What Comes Next

Meta says accuracy improved as the amount of training data grew, which suggests more data could push performance higher. The company also said AI agents explored possible optimisations for the decoding pipeline before its engineers chose the final configuration. Meta published a paper in Nature Neuroscience and is releasing the code and dataset through its Digital Brain Project, framing the work as open research meant to speed up how the field identifies, diagnoses, and treats neurological disorders.

What This Means in Practice

Brain-computer interfaces are moving from a surgical frontier toward something closer to a wearable. That shift matters for accessibility first, since it could restore communication for people with paralysis or brain injury without an operation. It also hints at where consumer neurotech may head, because the same techniques that decode intended text from brain signals could eventually inform hands-free interfaces. For now the practical takeaway is that non-invasive decoding has crossed from near-random accuracy into genuinely useful territory, and it is being released in the open for other researchers to build on.

The Bigger Picture

The announcement lands in a crowded and fast-moving field. Elon Musk’s Neuralink and Synchron are pursuing implanted interfaces, Merge Labs, backed by OpenAI chief Sam Altman, is developing its own technology, and startups are racing to improve non-invasive systems. Neurable introduced AI-powered EEG headphones in 2024, and MIT spinout AlterEgo unveiled a wearable that turns silent signals from the face and throat into text. Meta’s contribution is to push non-invasive accuracy toward what once required surgery, and to do it in the open. If that trajectory holds, the question is no longer whether thought can be turned into text, but how soon it can be done comfortably, accurately, and without an operation.

Frequently Asked Questions