Four products have been accepted into the FDA’s TEMPO pilot program, including AI tools from Cadence and Limbic. The program allows digital health companies to release their devices without the standard clearance pathway, giving regulators and developers a controlled way to study how generative AI behaves in real clinical settings.
Why the FDA is moving now
Generative AI in medicine poses a problem traditional device reviews were never built for. The models can update, drift, and produce novel outputs over time, which makes the static “review it once and approve it” model a poor fit. By allowing limited launches through TEMPO, the FDA gains real-world evidence on safety and performance that lab submissions cannot provide.
The pilot is also tied to a specific policy goal. TEMPO is designed to expand the catalogue of technologies available under the Medicare ACCESS model, an experiment in paying for digital tools that help beneficiaries manage chronic conditions. Without enough qualifying devices, that payment experiment would have little to test.
The companies in the program
Cadence and Limbic are the two named participants, with two additional devices also accepted. Both companies sit in the growing digital therapeutics space, where software is positioned as a clinical intervention rather than a convenience tool. For these firms, TEMPO offers something scarce in FDA device regulation: a route to patients that does not require waiting out a full authorization cycle.
That access comes with tradeoffs. Products launched under the pilot are not cleared, which means payers, clinicians, and patients must weigh evidence that is still accumulating. The arrangement also places a heavier monitoring burden on developers, who must generate the data that could eventually support a full submission.
AI regulation meets real-world care
Beyond Medicare ACCESS, the TEMPO pilot doubles as a regulatory laboratory. The FDA has struggled to define how it will oversee generative AI that continuously changes as it is trained or fine-tuned. Letting devices operate under provisional terms gives the agency a chance to observe failure modes, usage patterns, and clinical impact at scale before committing to binding rules.
That approach carries risk. If a provisionally launched device causes harm, the FDA will face questions about whether the pilot’s guardrails were adequate. The agency is betting that the pace of AI development makes that risk preferable to the alternative: regulating a technology it does not yet understand.
What Happens Next
Watch for two signals in the coming months. First, whether the FDA publishes performance or safety data from the TEMPO devices, which would set expectations for how transparent future AI pilots must be. Second, whether Medicare ACCESS adoption picks up enough to justify the pilot’s existence, or whether the program gets folded into a broader AI regulatory framework before it matures. The four accepted devices are effectively the test case for how the FDA handles generative AI medical devices from here — and how much latitude developers can expect before full authorization becomes unavoidable.
— Aisha Mensah, health desk, AXO News