For a long time, the idea of directly converting a thought into text belonged to the realm of science fiction. Brain-machine interfaces were limited to specialized laboratories or relied on brain implants that required surgery. With Brain2Qwerty v2, Meta is taking a new step forward by introducing artificial intelligence capable of interpreting brain activity and converting it into text without any brain implants.1 This breakthrough does not mean that the company is now capable of “reading minds” in the conventional sense. However, it does demonstrate that it is becoming possible to identify certain language-related intentions based on brain signals, thereby opening up new possibilities for medicine, neuroscience, and future interactions between humans and artificial intelligence.
\n\n\n\nBrain2Qwerty v2 is part of Meta’s research into non-invasive brain-computer interfaces, which has been ongoing for several years. Unlike solutions developed by companies such as Neuralink, which rely on implanting electrodes directly into the brain, Meta’s technology operates exclusively using an external magnetoencephalography (MEG) device. This equipment is capable of recording the very weak magnetic fields naturally produced by the electrical activity of neurons while a person types on a keyboard.1
\n\n\n\nThe concept is particularly innovative. The artificial intelligence does not seek to track finger movements or keystrokes. Instead, it directly analyzes the brain signals produced when the brain prepares to write a word or a sentence. By combining this information with deep learning models, the system then attempts to reconstruct the text the user intends to produce. This approach is gradually bringing brain-machine interfaces closer to more natural communication between humans and artificial intelligence systems.
\n\n\n\nThe first generation of Brain2Qwerty was primarily capable of recognizing characters one by one. This new version marks a significant milestone, as the system is now capable of reconstructing complete words and then entire sentences by leveraging linguistic context.2
\n\n\n\nTo achieve this, Meta combines data from brain scans with the power of large language models. The process is similar to predictive text on smartphones: when some information is missing or difficult to interpret, the AI uses the overall context of the sentence to estimate the most likely words. This strategy significantly improves decoding quality, even when brain signals are particularly weak and noisy.
\n\n\n\nIn the background, several deep learning technologies operate simultaneously. Convolutional neural networks analyze signals from the MEG scanner, while Transformer-based architectures interpret the complex time series generated by brain activity. These results are then enhanced by language models that act as intelligent correctors capable of reconstructing coherent sentences even when some information is incomplete. Meta also notes that it used artificial intelligence agents to automatically optimize the decoding process and progressively improve the system’s performance.1
\n\n\n\nTo train Brain2Qwerty v2, Meta built a particularly large database. Nine volunteers participated in the experiments, each spending nearly ten hours in an MEG scanner to record their brain activity while typing. In total, approximately 22,000 sentences were collected to enable the artificial intelligence models to learn to associate specific neural signatures with precise writing intentions.1
\n\n\n\nThe initial results appear particularly encouraging. According to Meta, Brain2Qwerty v2 achieves an average accuracy of 61% in word recognition. For the top performer, this accuracy reaches 78%, while more than half of the decoded sentences contain no more than a single word error. Although this performance is still insufficient for everyday use, it demonstrates that non-invasive brain-machine interfaces are advancing rapidly and becoming capable of processing increasingly complex linguistic information.
\n\n\n\nMeta’s approach differs significantly from that of Neuralink. While the company founded by Elon Musk relies on implants placed directly in the brain to achieve maximum precision, Meta favors a completely non-invasive solution. No sensors are inserted into the brain, and no surgery is required to use Brain2Qwerty v2.3
\n\n\n\nThis strategy offers several advantages in terms of medical safety, but it also comes with significant limitations. The MEG scanners used today are extremely bulky and expensive, and are found exclusively in certain specialized laboratories or hospitals. We are therefore still a long way from a device that anyone could use at home to compose a message simply by thinking.
\n\n\n\nDespite these limitations, researchers believe that the progress made in recent years shows that noninvasive brain-machine interfaces are gradually becoming a credible alternative to implanted solutions, particularly for certain medical applications.
\n\n\n\nAlthough Brain2Qwerty v2 will not be marketed to the general public in the near future, its potential in the medical field is considerable. People with amyotrophic lateral sclerosis (ALS), stroke survivors, and those suffering from severe paralysis could regain the ability to communicate thanks to this type of technology.
\n\n\n\nThe ability to directly convert a thought into text would represent a major breakthrough for these patients, allowing them to communicate without using either their voice or muscle movements. Beyond this application, Brain2Qwerty v2 could also contribute to a better understanding of how language functions in the human brain and accelerate research into future brain-computer interfaces.
\n\n\n\nAs part of an open research initiative, Meta has chosen to release some of the training code and the datasets used as open source, to enable other research teams to continue this work and improve the performance of future generations of models.1
\n\n\n\nEven though Brain2Qwerty v2 does not allow for the unrestricted reading of a person’s thoughts, this breakthrough is already raising major questions. Brain data is among the most sensitive information a human being can produce. As these models continue to advance, protecting this new category of data will become a critical issue.
\n\n\n\nIn particular, researchers will need to address several questions: How can we ensure that this information remains confidential? Who will have access to the collected brain signals? Under what conditions will a user be able to give or withdraw consent? Finally, how can we prevent future brain-machine interfaces from being used for surveillance or manipulation?
\n\n\n\nAs with many innovations in artificial intelligence, technological advances will need to be accompanied by in-depth consideration of the legal frameworks, ethical principles, and governance of these systems in order to protect individuals’ cognitive freedom and privacy.
\n\n\n\nWith Brain2Qwerty v2, Meta is not yet offering a product intended for the general public. However, the company is demonstrating that non-surgical brain-computer interfaces are advancing rapidly and that artificial intelligence now plays a central role in their development. The combination of neuroscience, deep learning, and language models opens up unprecedented possibilities for medicine, scientific research, and future interactions between humans and machines.
\n\n\n\nWhile many technical challenges still need to be overcome before this technology can be used on a large scale, this breakthrough confirms that the next revolution in artificial intelligence may no longer rely solely on screens, keyboards, or voice commands. It could begin right inside our brains.
\n\n\n\n\n\n\n Brain2Qwerty v2 is an artificial intelligence system developed by Meta to convert brain activity into text without the need for an intracranial implant. Unlike invasive brain-machine interfaces that require surgery, this technology relies on a completely non-invasive approach using magnetoencephalography (MEG), a functional imaging technique capable of measuring the very weak magnetic fields generated by the electrical activity of neurons.\n
\n\n\n When a user types text on a keyboard, the MEG scanner records, in real time, the brain signals associated with the preparation of movements and speech production. These signals are then analyzed by several specialized artificial intelligence models that attempt to reconstruct the words and sentences the person intends to write. The goal is not to “read minds” in the strict sense, but to identify the intentions related to language production based on the information emitted by the brain.\n
\n\n\n To improve decoding accuracy, Brain2Qwerty v2 combines several deep learning techniques. Convolutional neural networks (CNNs) extract the most relevant features from brain signals, while Transformer-based architectures analyze the temporal relationships between different neural activities. Finally, large language models act as a contextual correction system capable of filling in missing information and reconstructing coherent sentences, much like the predictive text tools found on smartphones. This second generation can now interpret words, expressions, and complete sentences while maintaining a completely non-invasive architecture.\n
\n\nMeta’s work on Brain2Qwerty v2 shows that artificial intelligence is gradually making its mark in neuroscience and next-generation medical technologies. On a related topic, check out our article “Scanning an Entire Body in 60 Seconds: Midjourney Unveils a 3D Scanner Up to 100 Times Faster Than an MRI , ” which explores how AI is also transforming medical imaging and opening up new possibilities for diagnosis and prevention.
\n\n\n\n1. Meta AI. (2026). Brain2Qwerty v2: Advancing Non-Invasive Brain-to-Text Interfaces.
https://ai.meta.com/research/
2. Meta AI. (2026). Brain2Qwerty v2 Open Research Repository.
https://github.com/facebookresearch
3. Neuralink. (2026). Brain-Computer Interface Research.
https://neuralink.com
For a long time, the idea of directly converting a thought into text belonged to the realm of science fiction. Brain-machine interfaces were limited to specialized laboratories or relied on brain implants that required surgery….
\n","protected":false},"author":7,"featured_media":658276,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"content-type":"","footnotes":""},"categories":[27],"tags":[59],"class_list":["post-658275","post","type-post","status-publish","format-standard","has-post-thumbnail","category-sante","tag-parlonsia"],"yoast_head":"\n