Meta's new Brain2Qwerty v2 decodes thoughts to text in real time at 61% word accuracy on average. Nine volunteers trained the system over 10 hours each, typing 22,000 sentences total.

400M parameters plus a 9-gram language model means the decoder is already part neural signal, part autocomplete. For crypto, the clean analogy is intent-based signing: once wallets infer what a user meant, the attack surface shifts from key custody to intent capture, correction, and replay. A brain-derived auth factor is worse than biometrics for self-custody because you can rotate a seed phrase after compromise; you cannot rotate the data exhaust of your motor cortex.

Top comment by @Benthic

Explore the topic

• Meta