Skip to content
See the World Through ScienceA project of ALLATRA
Source: Peer-reviewedNature Medicine1 source

Epilepsy Patients Asked an AI What Their Brain Monitor Saw, and It Answered

Science

Republish this story

Our work is licensed under Creative Commons BY-NC 4.0. You may republish this piece for free — with credit to ALLATRA Media and a link to the original, unedited beyond length trims, and not for commercial use.

Read the full license

A man wearing a dry-electrode EEG headset, with cables running from sensors pressed against his scalp.
A wearable EEG headset picks up the brain's electrical activity through the scalp. The hospital platform tested at Penn answered patients' questions from recordings of that same signal (illustrative)."Quantitative Electroencephalography (QEEG)" by SHA MASTERS OF LONGEVITY, via Wikimedia, CC BY-SA 4.0

A group at the University of Pennsylvania has built a system that lets people under hospital seizure monitoring message an AI from a smartphone and ask what their own brain recording is showing, and has tested it in 13 patients. The work was published Oct. 13, 2026, in Nature Medicine, which labels it open access.

The authors describe the platform as closing a gap in what a medical device tells the person wearing it. The Nature Medicine paper states that monitors can already flag an oncoming seizure, a missed dose of medication or a dip in attention, but that they pass it to the clinic rather than to the patient. The group calls the platform a proof of concept, makes no claim about an effect on seizures, and says a study in outpatients is planned.

A two-part schematic: smartphone chat screens showing seizure alerts and charts on the left, and on the right the cloud pipeline that extracts brain measures and feeds an AI system.
Panel a is the patient's view, where alerts, surveys and charts of the recording arrive in the chat. Panel b is the cloud pipeline behind it, which turns the raw signal into the measures the AI is allowed to query. Fig. 1 from Zack Goldblum, Haoer Shi, Zhongchuan Xu, William K. S. Ojemann, Carlos A. Aguila, Kevin Long, Kevin Xie, Kerry C. Nix, Katherine G. Walsh, Ellie Chang, Sarah B. Lavelle, Brandon Bach, Kathryn A. Davis, Nishant Sinha, Lauren H. Hammer, Erin C. Conrad, Brian Litt (2026), "Human–artificial intelligence collaboration for comanaging epilepsy in real time", Nature Medicine. CC BY-NC-ND 4.0, resized

Zack Goldblum, Brian Litt and colleagues enrolled adults admitted to the university's epilepsy monitoring unit for continuous video and EEG monitoring (the electrical recording of brain activity) without changing their treatment. The recordings streamed to a secure cloud service that estimated seizure probability, sleep stage and medication load. A version of OpenAI's GPT-4o-mini running under US medical privacy rules answered patients' questions about those measures, sent alerts and ran short daily surveys. Patients and the system exchanged 1,307 messages, most of them ordinary conversation rather than requests for data.

Patients replied to most of the real-time seizure alerts the system sent them and marked 37.9% of those as false alarms, the kind of label passive monitoring never collects. In a simulated replay of the recordings rather than a live test, the authors report that retraining the detector on patients' answers alone cut its false-alarm rate by about 80%.

The paper states that the Food and Drug Administration has approved three devices to treat epilepsy, and that none of them shares what it records with patients in real time. About a third of the world's 70 million people with epilepsy keep having seizures despite medication.

The authors report that five of the patients asked for access to the platform once they were discharged.

Sources

Medical Disclaimer: This content is provided for general informational and educational purposes only. It is not medical advice and should not be used as a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare professional regarding any medical condition, concern, symptom, or treatment decision.

Spot an error?

Spot an error?

Report an error

Spotted a mistake on this page? Tell us what's wrong and our editors will take a look.

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We correct mistakes openly. Select any text to flag it. Fixes are logged under our Corrections Policy.

Report an error

Reporting on

Epilepsy Patients Asked an AI What Their Brain Monitor Saw, and It Answered

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We read every report. Corrections are logged publicly.