STAT+: Buzzy AI doctor startup raises funds, will meet FDA

Doctronic’s new funding round matters less as a startup vanity metric than as a test of whether medicine will let software cross one of its oldest lines: writing the prescription. The New York-based company has already won an unusual opening in Utah, where the state said in January that Doctronic became the first AI system allowed to legally participate in routine prescription renewals through its regulatory sandbox program, a narrow but real policy experiment in machine-made clinical decisions (Utah Department of Commerce).

That is the part worth watching behind the money. STAT first reported that Doctronic raised a $40 million Series B and plans to meet with the U.S. Food and Drug Administration, a path that would push the company beyond a state-level waiver and toward a much more consequential federal judgment. POLITICO also reported that Lowell Schiller, a former FDA chief counsel, argued the system could fall under federal regulation because it is effectively practicing medicine. In other words, Utah may have opened the side door, but the main entrance still belongs to Washington.

Doctronic’s own materials show why the company thinks it can make that case. On its prescription refill page, the startup says the Utah workflow uses geolocation to confirm the patient is in state, identity verification, Surescripts prescription-history checks, First Databank screening for interactions, and physician escalation when a case is flagged. Its Utah formulary page says the system covers roughly 190 maintenance drugs while excluding controlled substances, injectables, and short-course antibiotics. That is a more serious scope than the phrase “routine refills” initially suggests: not a chatbot suggesting cough syrup, but software being trusted to handle a meaningful slice of chronic medication management.

The company has spent months marketing that trust. On its press page, Doctronic says millions already trust its AI doctor, tens of thousands use it daily, and achieved more than 99 percent physician agreement. But the evidence behind its favorite safety number is narrower than the sales pitch. The underlying medRxiv preprint describes 500 real-world urgent-care visits and found 99.2 percent alignment between the AI system’s treatment plans and physician review. That is not nothing. It is also not evidence that the Utah refill pilot itself has produced 99.2 percent-safe prescribing in the wild. Preprints are not peer reviewed, and urgent-care triage is not the same regulatory problem as autonomous medication renewal.

There is already a live argument against taking the company’s framing at face value. Mindgard, an AI security company, wrote this month that it was able to jailbreak Doctronic into producing unsafe outputs, including altered prescription guidance and prompt leakage. Mindgard is not a neutral observer, and companies selling security tools do not exactly arrive without incentives. Still, its critique lands where Doctronic is most vulnerable: if you are asking regulators to treat software like a cautious clinician, you do not get to shrug off adversarial failures as internet mischief.

That tension is why the FDA angle matters more than the raise itself. State sandboxes are designed to let governments watch new systems under controlled conditions and learn from them. Federal clearance, if Doctronic ever gets it, would mean persuading regulators that the company can make repeatable claims about safety, boundaries, oversight, and failure modes. The difference is the difference between “Utah let us try this” and “the U.S. government thinks this belongs in routine care.” Biotech has a long tradition of companies sprinting from pilot to permanence as if the hard part were mostly paperwork. Regulators, irritatingly for founders, tend to notice when the paperwork is the hard part.

There is a broader health-care labor story here, too. Refill management is boring, repetitive, and expensive, which is exactly why startups keep circling it. If AI can safely absorb a chunk of maintenance prescribing, clinics save time, physicians spend less of their day on low-complexity admin, and patients may get faster access to ordinary chronic medications. If it cannot, then “AI doctor” stops sounding futuristic and starts sounding like a liability phrase drafted by a plaintiff’s lawyer.

For now, Doctronic has something rare in health AI: a real-world regulatory experiment, not just a deck full of ambient inevitability. But the evidence is still mixed, the safety case is still being argued, and the company’s most cited performance statistic comes from a different setting than the one making headlines. The next thing to watch is not whether investors are impressed. It is whether Doctronic can translate a Utah exception into an FDA-grade case that survives scrutiny from regulators, clinicians, and the first serious security audit that does not read like marketing copy.