FDA to open a unit of 13 engineers including devs, experts in AI and cloud computing, to keep pace with changes to medical devices powered by machine learning
FDA's Bakul Patel envisions a new regulatory approach to digital health
Context & Ripple Effects
In 2017, the FDA's answer to machine-learning medical devices was a single unit: 13 engineers, including developers and cloud specialists, assembled under Bakul Patel to write a new regulatory approach to digital health. The bet was that reviewers needed in-house technical fluency before they could evaluate products that change after clearance.
That bet compounded. By 2025 the agency had named its first AI chief and moved to deploy AI tools across all centers, then debuted an agencywide generative AI tool for scientific reviewers. By 2026 it was running an AI-and-cloud pilot giving regulators a direct feed of real-time clinical trial data, and the White House was pushing an FDA fast track for digital health tech like AI chatbots. The 13-person unit is the origin point of that arc.
First-order effects
- Device makers shipping ML-powered products gain reviewers who can read code and cloud architecture, so adaptive algorithms stop being evaluated with static-device templates that don't fit them.
- Bakul Patel's team becomes the drafting floor for how the FDA treats software that updates post-market, setting the terms every digital-health applicant must meet.
Second-order effects
- Consumer hardware companies get a workable path into regulated health features — the clearances that followed for Apple Watch sleep apnea detection, Samsung's earlier version, Google's Loss of Pulse Detection on Pixel Watch 3, and AirPods Pro 2 as the first OTC hearing-aid software all route through this regulatory machinery.
- Rival regulators and other FDA centers face pressure to build equivalent engineering capacity rather than rely on sponsor-submitted evidence, since the review bottleneck moves from domain knowledge to software literacy.
Third-order effects
- If the pattern holds, medical-device regulation shifts from point-in-time clearance to continuous oversight — the logical endpoint visible in the 2026 direct-data-feed pilot, where the regulator watches trial data stream in rather than waiting for submissions.
- A regulator staffed with its own engineers becomes an active participant in shaping the AI healthcare market, not just a gatekeeper — which is exactly the posture the later fast-track push institutionalizes.
The trend: Health regulators are converting themselves into software organizations — starting with small embedded engineering units and scaling toward AI-native review infrastructure and expedited pathways for algorithmic medicine.