Sleep wearables are training AI on your brainwaves without explicit consent

Consumer sleep wearables are collecting EEG recordings that companies are using to train AI models, often without clearly telling users in advance. An investigation found that several neurotechnology firms have amassed tens of millions of hours of brain-wave data. Interaxon, maker of the Muse headband, has trained a model on roughly 50,000 hours of recordings and says it removes identifying details. Its privacy policy did not initially spell out this use; after questions were raised, the company updated the policy and agreed users should be explicitly informed. Emotiv also revised its policy to disclose model training with de-identified data, requiring opt-in consent. More than two-thirds of the companies examined reportedly permitted corporate transfers of user data without consent, raising concerns that acquisitions could place stored neural recordings under new ownership. California and Colorado have added protections for neural data, but coverage remains uneven.

Brain-data models learn patterns in electrical signals and can be adapted for tasks such as recognizing brain states; they do not make ordinary consumer headbands capable of reading thoughts. Such devices typically have only one to six electrodes, compared with roughly 19 to 25 in clinical EEG systems, and produce less complete recordings. The technology nevertheless has demonstrated potential: a model trained on more than four million EEG hours by AI startup Piramidal and Cleveland Clinic identified nine out of ten seizures in a pilot after clinical fine-tuning. Security is another risk: a smart sleep mask was found to expose other users’ EEG readings through credentials embedded in its app. These developments make consent, data security, and control over neural information central concerns as brain-sensing products spread.