FDA TEMPO Pilot Allows Pre-Approval Use of Generative AI Medical Devices

For digital health developers, the traditional regulatory pathway is a bottleneck that keeps potentially life-saving artificial intelligence tools out of patients' hands for years while agencies deliberate [1]. For the Food and Drug Administration, generative AI presents an unprecedented risk profile—models that can hallucinate, drift, and make autonomous care decisions cannot simply be waved through without rigorous, settled safety frameworks [1][4].[1][4]

The FDA is now attempting to bridge that divide through a regulatory compromise. Under the Technology-Enabled Meaningful Patient Outcomes (TEMPO) pilot, the agency is provisionally allowing select medical devices powered by generative AI to reach the market without formal premarket authorization [2]. The program grants "enforcement discretion" to a limited cohort of manufacturers, permitting them to deploy their tools in real-world clinical settings while the FDA gathers the data necessary to figure out how to regulate them permanently [4].[2][4]

The pilot is directly tethered to the Advancing Chronic Care with Effective, Scalable Solutions (ACCESS) model run by the Centers for Medicare and Medicaid Services (CMS) [3]. Launched in July 2026, the ACCESS model is an experiment in paying for technology that helps Medicare beneficiaries manage chronic conditions [3]. By linking TEMPO to ACCESS, the FDA ensures these AI tools are deployed within a structured reimbursement environment, targeting cardiometabolic risks, chronic pain, and behavioral health [3][4].[3][4]

As of late August 2026, the FDA has accepted four companies into the TEMPO pilot, with two explicitly deploying generative AI [3]. Limbic is fielding a product called Unpacked, which utilizes an AI voice agent to deliver cognitive behavioral therapy to Medicare patients suffering from significant depression or anxiety [3][5]. Cadence Solutions has introduced HypertensionOS, a tool designed to help physicians and advanced practice registered nurses adjust blood pressure medications for adults with Stage 2 hypertension [3].[3][5]

The cohort also includes Dexcom, which was the first participant announced in July 2026 [3]. Dexcom's Glucose Health Program pairs real-time glucose data with AI-driven insights to screen for prediabetes and Type 2 diabetes [3]. SonderMind rounds out the initial group with a smartphone application intended to assist adults 22 and older with moderate depression or anxiety, designed to be used alongside traditional therapy or medication [3].[3]

The core trade-off of the TEMPO program is explicit: earlier commercial access in exchange for rigorous, continuous surveillance [5]. By exercising enforcement discretion, the FDA allows device manufacturers to forego certain marketing requirements, such as labeling and premarket authorization, while participating in the pilot [4]. In return, participating manufacturers are required to collect, monitor, and report real-world performance data on their devices' intended uses, functioning as a continuous clinical study infrastructure [3][5].[3][4][5]

Generative AI poses unique challenges for medical device regulation because its outputs are not strictly deterministic. While the FDA has authorized more than 1,500 devices with an AI component so far, it had not authorized a generative AI-enabled device prior to this pilot [1]. In a discussion paper released in August 2026, the FDA acknowledged that generative AI devices may have a range of possible inputs and outputs that are too large to test comprehensively before market entry [1].[1]

"We're moving past a place where you can really see completely under the hood," said Suzanne Levy Friedman, a partner at Honigman, highlighting the agency's shift toward benchmarking and competency-based assessments [1]. "They're really starting to go through the motions more than just talk," she added, noting that the agency is actively soliciting comments on its generative AI discussion paper through mid-October 2026 [1][5].[1][5]

The FDA has made it clear that while TEMPO accelerates access, it does not equate to a declaration of proven efficacy [5]. The agency explicitly notes that the effectiveness of the devices selected for the pilot has not yet been evaluated for their intended uses [5]. Patients and caregivers are urged to understand the distinction between pilot access under enforcement discretion and full FDA clearance, particularly given the technology's known tendency to produce incorrect outputs [1][5].[1][5]

Looking ahead, the FDA plans to select up to 40 TEMPO pilot participants across the four clinical use areas, expanding the testing ground significantly [3]. For the broader medical technology industry, the pilot serves as a critical signal of how the agency intends to handle adaptive software [4]. The real-world data generated by Limbic, Cadence, and others over the coming months will likely form the bedrock of the FDA's eventual, formalized regulatory framework for generative AI in healthcare [4].[3][4]