Utah Opens Its AI Health Sandbox to Prescription Software and Women's Health Tools

Key points

• Utah's healthcare AI sandbox, a controlled testing programme that lets companies run AI health tools on real patients before full regulation kicks in, has expanded with new pilots approved in 2025.
• August AI, a company building software that generates prescriptions using artificial intelligence, is one of the newly approved participants.
• Nolla Health, focused on women's health, has also joined the sandbox.
• Utah has added third-party evaluators, independent reviewers not employed by the state or the companies being tested, to assess whether the pilots are safe and effective.
• The sandbox sits inside a broader national conversation about how to regulate AI in medicine before patients are harmed.

Utah is one of the few places in the United States where an AI company can legally run its health product on real patients under a supervised trial that sits outside the normal approval process. The state calls it a sandbox, borrowing a term from fintech, where it means a walled-off space for testing that can't spread to the wider world if something goes wrong.

What are these companies actually doing?

August AI's product is the one that'll raise eyebrows. The company describes its software as capable of generating AI-powered prescriptions, meaning the system recommends or drafts medication orders rather than a doctor typing them out. Prescriptions carry legal and clinical weight, and handing any part of that process to software is exactly the kind of thing regulators normally scrutinise for years before allowing near a patient.

Nolla Health's focus is women's health, though the specific tools it's piloting haven't been described in detail in the approvals made public so far. It's worth pairing this story with our 27 August report "AI reads mammograms for heart disease, not just breast cancer", which showed how AI is already finding new uses in routine women's health scans.

Both companies are operating under Utah's sandbox rules: they can collect real-world data from real patients, but within defined limits and under state oversight.

Why does adding third-party evaluators matter?

Until now, the main check on whether a sandbox pilot was working was largely the company's own reporting back to the state. Third-party evaluators change that. They're independent, they're not paid by the company being assessed, and their job is to look at the evidence and say plainly whether the tool is doing what it claims.

That's the piece of the structure that was most obviously missing. A programme that relies mainly on self-reporting from companies hoping to graduate into full commercial use isn't much of a safety net. Utah is trying to build a model other states might copy: get AI into clinical settings early enough to learn from it, but with enough independent scrutiny that problems surface before they scale.

Should patients in Utah be concerned?

Cautious is the right word, not alarmed. Sandbox participation doesn't mean a product is approved or proven safe. It means the state has decided the potential benefit justifies a structured trial.

Patients in any setting where these tools are being used should be told they're part of a pilot programme. If your provider mentions AI-assisted prescribing, ask whether that system is in a regulatory trial and what your right to opt out looks like. Watch for any provider who can't tell you whether an AI tool was involved in your care, any prescription recommendation that arrives without a human clinician reviewing and signing off, and any consent form that buries sandbox participation in small print. Our coverage of AI consultation notes making a GP's job harder is a useful reminder that even well-intentioned AI tools can introduce new risks at the clinical front line.