Your headband may have trained AI on your brain waves

Headbands and other wearables that measure the brain's electrical activity are used for meditation and sleep. But the data can also end up as training material for artificial intelligence. That is the finding of a review in the journal Science, as reported by the news site Inside BCI.

The measurements are called EEG. They are small electrical signals from the brain that sensors on the head pick up.

Muse trained AI on user data

According to Science, the Canadian maker Interaxon, which is behind the Muse headband, trained a brain AI model on roughly 50,000 hours of EEG data from users. The study says the data was stripped of names and emails, and the company plans to use more data.

Muse's privacy policy originally did not mention that brain data is used to train AI. After inquiries from Science, Interaxon changed the policy, and it now states that de-identified data is used to train machine learning models. De-identified means that information pointing directly to a person has been removed.

Interaxon's marketing chief told Science, according to Inside BCI, that brain data is the most valuable part of the product.

In June 2025, Muse announced a so-called "foundation brain model." It is built on the same type of technology as ChatGPT and, according to IEEE Spectrum, was trained on more than a billion minutes of brain data from Muse users over ten years. The sources do not make clear whether it is the same model and the same dataset as in the Science study. The figures are hard to reconcile.

Interaxon co-founder Chris Aimone says: "You have to train something that has the complexity of a brain in order to understand a brain."

Weak rules across the industry

Science reviewed the privacy policies of 57 consumer-facing EEG companies:

- 4 had no policy. - 11 had a policy only for visitors to their website. - Only 6 explicitly stated that they do not sell brain data. - All but one of the companies that have a privacy policy allow sharing with third parties in anonymized, aggregated or de-identified form. - More than two thirds allow data to be transferred to a new owner in an acquisition.

According to Inside BCI, Emotiv trains a "Large Brain Model" partly on data from users' devices. The company has changed its policy effective August 6, 2026, so that explicit consent (opt-in) is required.

Following inquiries from Science, Interaxon has changed Muse's policy so that it now says de-identified data is used to train machine learning models, including foundation models. Explicit consent (opt-in) applies to Emotiv's new policy, not Muse's.

Legislation has begun to respond

In the US, Connecticut has classified neural data as sensitive personal information as of July 1, 2026. California and Colorado have also included neural data in their privacy laws.