On October 14, 2026, Meta unveiled ten research manuscripts produced with help from its Muse AI platform. All ten draw on electronic health records, and all ten were carried out with human experts supervising the work.
Meta’s system tries to automate research preparation. The work centers on Muse Spark, a component of the broader Muse platform, which handles four core tasks: identifying patient cohorts, running statistical analysis, retrieving relevant literature, and drafting manuscripts.
The workflow relies on multi-agent AI models. That means several AI components split up the job, each handling a different piece, while human experts oversee the process.
The underlying dataset covers detailed health information on 4.95 million patients, drawn from an EHR release in June 2026. Within that pool, 156,051 patients carry breast cancer codes.
One of the ten studies chased a genuinely obscure target: Cladophialophora bantiana, a rare fungal pathogen. The project was prompted by a report in the Morbidity and Mortality Weekly Report, or MMWR, which flagged a rise in cases.
Using the dataset, the study identified 21 confirmed or probable infection cases. Seven of those patients had brain infections. Two of the seven died.
Another notable study looked at breast cancer patients who received antibody-drug conjugates one after another. The study examined two of these drugs, abbreviated SG and T-DXd, and the order in which patients received them.
When patients got SG after T-DXd, the median interval between treatments was 63 days. In the reverse order, the median interval was 49 days.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.