For decades, there has been an unwritten rule in health technology: software can assist, but it cannot decide.
AI is being used to transcribe doctor notes, suggest billing codes, or surface clinical guidelines. But the final decision to authorise a prescription has remained strictly with a licensed practitioner.
That line has now been crossed.
The state of Utah has officially authorised Doctronic, an AI-native healthcare platform, to act as an autonomous clinical decision-maker.
Not as a tool for doctors. Not as a second opinion. But as a system allowed to evaluate patients and authorise medication refills on its own.
It’s a quiet regulatory move with potentially far-reaching consequences. It also signals how states, not federal agencies, may shape the next phase of AI in healthcare.
Utah-Doctronic Partnership: What has Utah approved?
The collaboration is the flagship project of Utah’s Office of Artificial Intelligence Policy (OAIP) and its AI Regulatory Sandbox.
Established by the state legislature in 2024 through Senate Bill 149, the sandbox creates a “safe zone” where companies can test technologies that might otherwise conflict with existing state laws.
In this specific case, Utah has granted Doctronic a temporary waiver from certain statutes governing professional licensure and telehealth prescribing.
This allows Doctronic’s AI to evaluate and authorise medication refills without a doctor reviewing every individual request in real time.
The target problem is administrative overload, and the goal is to solve the “refill crisis.”
Primary care clinics across the US are drowning in refill requests for patients with stable, chronic conditions. Delays can stretch for weeks, leading to missed doses, worsening outcomes, and frustrated patients.
By delegating routine renewals to a qualified AI, Utah hopes to improve medication adherence and free up clinicians for more complex patient needs.
This is not “AI with a blank cheque”
Despite the headline, the system is tightly constrained.
The Doctronic platform functions within a highly structured environment defined by clinical guardrails and a phased oversight model.
The AI is authorised to work only within a defined list of 191 low-risk medications. These include:
- Statins for cholesterol management.
- Blood pressure medications (antihypertensives).
- Oral contraceptives.
- Certain stable psychiatric medications.
What’s excluded is just as important.
The system cannot prescribe controlled substances, narcotics, stimulants, antibiotics, or any short-term or high-risk drugs. It also cannot initiate new treatments. The AI can only approve refills for existing diagnoses and ongoing therapies.
How Utah is managing risk
To address safety concerns, Utah OAIP has mandated a phased rollout with escalating autonomy.
- Phase 1 (Calibration):
The first 250 refills processed by the AI must be reviewed and co-signed by a licensed clinician. This ensures the AI’s logic aligns with standard medical practice.
- Phase 2 (Auditing):
Once the initial threshold is met, the system enters a 10% audit phase. A doctor reviews one out of every ten decisions made by the AI.
- Phase 3 (Monitoring):
If the system maintains its safety record, it moves to random sampling and ongoing data monitoring.
How Doctronic’s AI works
Behind the scenes, Doctronic’s platform uses a multi-agent AI architecture. Instead of a single model making decisions, multiple agents work together.
- One checks the patient’s medical history,
- Another cross-references the latest clinical guidelines,
- and third acts as a “safety officer” to flag contraindications.
It’s meant to replicate the internal checks that happen in a physical clinic, but in software form.
Why this partnership is important for future AI deployments in health
The Utah-Doctronic partnership is a signal that the industry is moving from Assistive AI (the scribe) to Agentic AI (the practitioner). From tools that support clinicians to systems that perform clinical tasks themselves.
This shift has three major implications for the future of healthtech:
First, regulation is moving to the state level.
As federal regulation through the FDA remains slow to adapt to autonomous software, states like Utah are taking the lead as innovation hubs.
Utah’s “sandbox” model is already being mirrored in Arizona and considered in Texas.
We may see a future where medical AI is “licensed” at the state level, similar to how human doctors are, allowing for localised testing and faster iteration.
Second, policy is becoming data-driven.
The OAIP requires Doctronic to submit monthly reports detailing every denial, approval, error and clinician observation. This creates a “robust evidentiary case.”
Instead of debating the ethics of AI in the abstract, lawmakers will have hard data on whether an AI-driven refill is safer or more efficient than a human-driven one.
This data will likely form the basis for future national legislation.
Third, the burnout problem is being addressed head-on.
Perhaps most importantly, this pilot addresses the economic reality of modern medicine.
By automating the bottom 80% of routine clinical tasks, healthcare systems can redirect their most expensive and scarce human expertise towards patients who need it the most.
If successful, the Doctronic model proves that AI can replace the “busywork” that prevents doctors from practising medicine.
Wrapping up
Utah’s experiment is still a pilot. But it will serve as the ultimate proof-of-concept for autonomous clinical care.
If the data from Utah shows that AI can handle refills safely, the “red line” of clinical decision-making may be permanently moved.
-By Rohini Kundu and the AHT Team
