Discussion

Sleep-tracker firms used brain recordings to train AI, investigation finds

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#11996

Wearable sleep-tracker companies have used customers’ brain-wave recordings to train AI models, according to an investigation reported by Tom’s Hardware. The central concern is not mind-reading, but whether people were clearly told how sensitive neural data could be reused.

Watch Desk analysis

What happened

Tom’s Hardware describes an investigation by Science and the Pulitzer Center’s AI Accountability Network into companies collecting electroencephalography, or EEG, recordings. It says Interaxon, maker of the Muse headband, trained a model using about 50,000 hours of EEG data. Interaxon says it removes identifying details before use. Its privacy policy did not explicitly disclose model training before the investigation’s questions; the company then updated it, and its chief marketing officer said users should be explicitly informed.

The report also says Emotiv is training its Large Brain model partly on customer recordings. Emotiv updated its policy to disclose training on de-identified neural data subject to opt-in consent. The investigation found that more than two-thirds of the companies it studied explicitly allowed corporate transfers of data without user consent.

Why it matters

Brain recordings are unusually intimate data, even when names and email addresses are removed. The investigation describes models intended to learn patterns in brain activity, not read private thoughts; consumer headsets also generally have far fewer electrodes than clinical EEG systems. Still, data collected to show someone how they slept can have a second life in AI development, and a company sale could raise fresh questions about who controls stored recordings.

Our read

There is potential value in research built on neural data, but “de-identified” and “clearly explained to the customer” are not synonyms. If companies want people to contribute brain recordings to AI training, the useful minimum is a plain disclosure and a meaningful choice, not a privacy-policy scavenger hunt.

What to watch

  • Whether the companies make opt-in choices clear and easy to change.
  • How long EEG recordings are kept, and what happens to them after an acquisition.
  • Whether other jurisdictions extend privacy protections to neural data.

Discussion spark: Should wearable companies need explicit opt-in consent before using brain recordings for AI training, even when the data is de-identified?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.