Doctronic’s raise impresses, but so do the questions around its AI tech. Earlier this week a hot AI telehealth startup announced a hefty (for these times) $40 million Series B raise that topped off a Series A of $20 million last September and a $5 million seed round in April. The $40 million was funded by Lightspeed Venture Partners, Union Square Ventures, MANTIS Venture Capital, Davidovs Venture Collective, and Abstract. The New York City-based company was founded in 2023 and only constructed its clinician network in early 2025. It claims to be on track to earn $10 million in revenue in 2026.

The basic health tech sounds not that unusual: a chatbot discusses your medical concerns and questions, much like a Claude for Healthcare, Microsoft Copilot Health, Teladoc, Ro, or even Google Gemini. The next step is a clinician referral, available 24/7 in all 50 states, for a low $39. It also claims to securely retain your information and timeline/meds/labs, not using the data for AI training.

Where the controversy centers is Doctronic’s first-ever state-approved autonomous AI test with the state of Utah. Announced in January, it will test whether a chatbot agent can evaluate and renew existing prescriptions for Utahns without human clinical oversight. In the pilot first phase, the renewals, which include 192 drugs for chronic conditions, will be overseen by clinicians before being sent to a pharmacy, but the intent is to move through this phase quickly to a pilot of full prescription renewal autonomy. Utah is permitting this through the Utah Department of Commerce’s Office of Artificial Intelligence Policy. The goal is to speed renewals of maintenance medications, the majority of activity, thus reducing medication noncompliance. Non-compliance is a leading driver of preventable health outcomes and with health decline, avoidable spending. Utah Department of Commerce release, Doctronic blog

Benefit manager Healthesystems outlined the process for its blog, with a view to the issues. “At the Doctronic prescription renewal portal, patients must confirm that they are located in Utah, enter the medication they want refilled, and then select an in-state pharmacy for fulfillment. Users must then upload their ID, along with a verification selfie and proof of an old prescription and then pay a $4 service fee. The AI system reviews the information to ensure a prescription history exists, after which a health assessment is given, where patients must answer certain questions before the program issues a refill. If the AI is uncertain if a prescription should be renewed, it refers the patient to a Utah-licensed human physician.”

A big reveal is here. The physician review prior to being sent to the pharmacy is for only the first 250 patients; the next 1,000 patients will be reviewed retrospectively. After that, only 5-10% of renewals will be audited after the fact, monthly. The issues for the two Healthesystems reviewers are risk–missing a loss of stability or extended renewals beyond original intent, for instance–and maintenance of oversight. Who is ultimately held accountable for the chatbot’s actions? 

A JAMA Health Forum article (19 March) raises additional issues. There apparently has been no pretesting of the prescription chatbot, only simulation testing. The application references a preprint study in medRxiv, written by equity owners in the company and only about the existing website chatbot. Moreover, the agreement is well-hedged to protect Doctronic. FTA:

If prescription errors injure patients, Doctronic’s accountability is murky. Its contract requires it to compensate Utah for any liability costs the state incurs and Doctronic took out a special malpractice insurance policy. Yet, the terms of service that users of the prescription renewal system must agree to—which seem to have been developed for the company’s AI doctor system—currently state that Doctronic disclaims all responsibility and liability for system accuracy or harmful outcomes.

A further oddity is that the Utah contract relieves Doctronic of the obligation to “generate, maintain, and make available to each patient” the patient’s medical records.

It closes with a short discussion of ‘scope creep’ (Editor’s emphasis): “Once an AI system has secured acceptance, vendors may be able to push updates that include substantial changes without attracting the same degree of scrutiny as the initial adoption. Concerns that low-risk pilot programs may legitimize higher-risk deployments at scale have been expressed about the Centers for Medicare & Medicaid Services’ new pilot program using AI to conduct prior authorization reviews of some services in traditional Medicare plans.”

And where is the proof that the AI chatbot can’t be spoofed? STAT (Mario Aguilar), who has been following Doctronic, located a February test by UK-based cybersecurity firm Mindgard that tested Doctronic’s existing chatbot and fooled it into into believing deliberate “official” misinformation, a bogus guideline that allowed triple the standard adult normal dose of Oxycontin, a Schedule II controlled substance. Sergei Polevikov in AI Health Uncut (subscription required, but you should) describes it in far greater and scary detail, including his own test. He also points out and analyzes other Doctronic questionable claims, such as volume (claiming 24 million ‘helped’ not borne out by website traffic), problems with the Utah formulary and refilling several problematic drugs, an odd connection with a Belarus company, and whether this should be regulated by FDA.  To be continued.