America is letting some AI health tools reach patients before approving them
What happened
The United States medicines and devices regulator, the Food and Drug Administration, has started a pilot scheme that allows a small number of artificial intelligence health products to be used by patients before the regulator has formally approved them. The scheme is called TEMPO, which stands for Technology-Enabled Meaningful Patient Outcomes. The FDA announced the first participant in July 2026. Reporting by the health news outlet STAT on 3 September 2026 covered the latest companies to be accepted. The FDA's published participant list names four manufacturers: SonderMind, Limbic, Cadence Solutions and Dexcom. Here is the unusual part. For these companies, the FDA says it intends to use what it calls enforcement discretion for certain requirements, including the usual requirement to get authorisation before going to market. In plain English: the regulator has said it will not stop them from launching while the evidence is still being gathered. It is not a free-for-all. The products must be used within a specific government health programme, run by the Medicare innovation centre, aimed at helping people manage long-term conditions. Limbic, which offers an AI voice service for mental health support, will collect real-world data on outcomes, safety and how patients actually use it, which will later support its formal application for approval.
Why this matters
Regulators are stuck with a genuine problem, and it is worth understanding it rather than picking a side straight away. The usual way to approve a medical device is to test a fixed thing. You define exactly what it does, you test it, you approve that version. Generative AI does not sit still like that. The models change. The answers vary. The same question asked twice may not produce identical words. So regulators face a choice. Wait until they have a perfect method, and hold back tools that might help people. Or let some tools out under controlled conditions and learn from what happens. TEMPO is the second option. The FDA gets hands-on experience of how these products behave with real patients, and patients in the programme get access sooner. The risk is equally obvious. Patients are using health tools whose benefit has not yet been proven. Limbic's participation gives it potential reach to a very large number of Medicare beneficiaries. That is a lot of people relying on something still being evaluated. For UK readers, this is not a change to anything you can access. It matters because regulators watch each other. What America learns here will inform the debate in Britain and Europe.
What the evidence actually says
The honest answer is that the evidence is exactly what this scheme is designed to produce, which means it does not exist yet. That is not a criticism. It is the point of the pilot. But it does mean nobody can currently tell you whether these AI tools work better than what patients would otherwise get. What we do know is the structure. Participation is limited to four named manufacturers. Use is tied to a specific care programme rather than open to anyone. Data on outcomes, safety and usage is being collected. And the companies still have to seek proper authorisation afterwards. Those are meaningful guardrails, and they are more than a simple free pass. But they are not the same as proof. It is also fair to note the criticism that has followed. Reporting on the scheme has raised the concern that people receiving AI-delivered talk therapy through this route are using a service before the regulator has evaluated whether it is effective. That is an accurate description of what is happening, and it is a reasonable thing to be uneasy about. The UK position is different. Here, the MHRA regulates medical devices, including software, and is currently working on how to handle AI products. Nothing described here applies to tools you might be offered by the NHS.
Practical advice
If you use a health app, and many people do, a few simple habits are worth having. Check who stands behind it. An app recommended by the NHS, or supplied through your GP practice, has been through a different process than one you found in an app store. Be careful about what you tell it. Health information is among the most sensitive data about you. Read what the app says it does with your data, and if that is not clear, treat that as an answer in itself. Treat an AI tool as something that helps you prepare, not something that decides. Writing down your symptoms before an appointment is useful. Concluding what is wrong with you from a chatbot is not. Never change or stop a prescribed medicine on the say-so of an app. And if an app tells you something worrying, do not sit with it alone. Ring your GP practice, or NHS 111 if it is urgent. Real clinicians would much rather hear from you than have you worry for a week.
What to know
The FDA is running a limited pilot, TEMPO, that lets four named companies put AI health products in front of patients before formal approval, within one specific care programme, while safety and outcome data is collected. It is a deliberate experiment in how to regulate technology that does not hold still, and it carries a real trade-off between faster access and unproven benefit. None of it changes what is available in the UK, where the MHRA sets the rules. But it is an early view of an argument that will reach the NHS soon enough. Sources: US Food and Drug Administration, 'Participants Selected for TEMPO for Digital Health Devices Pilot', 2026, https://www.fda.gov/medical-devices/digital-health-center-excellence/participants-selected-tempo-digital-health-devices-pilot ; STAT News, 'FDA pilot offers generative AI medical devices a path to patients before they are authorized', 3 September 2026, https://www.statnews.com/2026/09/03/tempo-fda-pilor-generative-ai-medical-device-regulation/ ; Fierce Healthcare, 'Limbic selected to FDA pilot, will test voice AI therapy with Medicare enrollees', 2026, https://www.fiercehealthcare.com/digital-health/limbic-selected-fda-tempo-track-outcomes-ai-therapy-medicare This article is for general information and does not replace advice from a doctor, pharmacist or other qualified healthcare professional. Health apps and AI tools are not a substitute for a clinical assessment.
Related articles
A knitted fabric that snaps into shape - and works as a switch
Engineers at Harvard have made knitted fabrics that can snap between different shapes and hold them, and that can double as soft electronic switches. The resea…
Scientists have turned bacteria into living transistors
Engineers at MIT have made bacteria behave like the components inside a computer. The work was published in Nature Chemical Biology and reported on 4 September…
An AI now steers fusion plasma faster than any human could
Researchers have built an artificial intelligence system that can steer the superheated gas inside a fusion reactor faster than a human operator could ever reac…