ChronicleBio, the chronic-disease startup Fidji Simo founded after stepping back from her OpenAI leadership role in July, will open a home phlebotomy signup on August 11. Mobile blood-draw trucks will visit patients at home in exchange for a personalized diagnostic report, free for the first 250 participants and $400, at cost, after that. Fortune’s Emily Forlini got Simo’s first interview since the departure, and most of it is about the company, not the exit.
Simo describes her health plainly rather than dramatically: she is, in her words, “physically the worst I’ve ever been.” She has lived with Postural Orthostatic Tachycardia Syndrome, or POTS, for seven years. There is no cure. The condition causes dizziness on standing, fatigue, brain fog, and headaches, and traces to an imbalance in the autonomic system that regulates heart rate and blood pressure.
The company’s actual bet has nothing to do with sympathy and everything to do with how clinical trials are structured. Drug trials group patients by symptom-based diagnosis, and Simo argues that a single diagnosis can conceal several distinct underlying diseases. If a drug works for only one hidden subgroup, testing it against the full cohort produces a null result. A drug that could help some patients gets shelved because it failed to help all of them.
ChronicleBio applied that logic to POTS and says it has already identified five biologically distinct sub-diseases sharing the same diagnosis. Patients report the same core symptom, a racing heart on standing, but the underlying drivers differ: in one subgroup the immune system appears to be the cause, in another it traces to mitochondrial function. Sub-diseases, in Simo’s framing, do not yet have names, because medical taxonomy classifies syndromes by symptom rather than mechanism.
That reclassification is the point of the whole exercise. ChronicleBio plans to test existing, already-approved drugs against each of the five subgroups before 2026 ends, betting that treatments abandoned as trial failures might work cleanly once matched to the right biological subgroup. A hit there would validate the subgrouping approach before ChronicleBio has to develop any new compound itself. The next phase, Simo says, involves bringing in biotech and pharmaceutical partners.
The scale behind the claim is still small but specific. In its first year, ChronicleBio drew blood 890 times from 709 patients spread across Utah, Arizona, Texas, and India, filling a biobank at its new Menlo Park lab with more than 3,500 tubes. That blood work has produced 153 terabytes of data, more than three times the 45 terabytes OpenAI used to train GPT-3.5, the 2022 model Forlini cites for comparison. Total funding to date stands at $15 million.
Simo’s explanation of why the company collects blood rather than relying on medical records is the sharpest idea in Forlini’s interview. Large language models worked, she told Fortune, because “the internet existed” and “you already had all of this language.” No equivalent corpus exists for human biology. “We are missing the internet of biology,” she said. Electronic health records, in her account, are noisy and organized around symptoms rather than the underlying interaction of the nervous, immune, and genetic systems. Building that missing corpus one blood draw at a time is what the August 11 launch is for.
The evidence for the five-subgroup claim is still internal to the company. ChronicleBio is a one-year-old startup with $15 million in funding and no published, peer-reviewed results validating the POTS subgrouping model Simo describes. Five biologically distinct sub-diseases is a hypothesis generated from ChronicleBio’s own data, not yet a finding that has survived outside review, and the approach only proves itself if the existing-drug tests due before year-end actually work.
The business model runs through consent as much as through blood chemistry. Participants trade biological data, including future retests that let ChronicleBio track disease progression over time, for a free diagnostic report. A biobank built from donated blood of chronically ill patients is a real asset for the company regardless of what the subgrouping model eventually proves, which raises its own questions about how a young, venture-funded startup governs years of longitudinal biological data from people who signed up for a free health report.
Operators watching AI’s application to biotech should track ChronicleBio’s year-end drug-matching results as the real test of the thesis, not the fundraising or the phlebotomy launch. If existing drugs succeed against a named subgroup where they failed against the broader diagnosis, the trial-design argument gets a data point that outside researchers can scrutinize. Until then, the five sub-diseases remain a claim from a company that stands to benefit from investors believing it.
This account draws on Fortune’s exclusive interview with Fidji Simo, reported by Emily Forlini and published August 3, 2026.
