FDA keeps AI devices in check
Washington rejected an industry bid to loosen oversight of certain AI medical devices, signalling regulators won't give AI in health an easy pass. That decision suggests founders whose products cross from wellness into clinical support should expect stricter review rather than lighter rules. (alltoc.com)
In January 2026, the Food and Drug Administration did not throw open the doors for health artificial intelligence. It released two final guidances that carved out more room for wellness gadgets and some doctor-facing software, but kept the agency in charge of anything that still looks like a medical device. (fda.gov) The split turns on a basic line: a wellness product helps you live healthier, while a medical device helps diagnose, treat, prevent, or mitigate disease. Congress drew that line in the 21st Century Cures Act in December 2016, and the Food and Drug Administration’s January 2026 wellness guidance says software aimed at a healthy lifestyle stays outside device rules only when it is unrelated to disease. (fda.gov) That is why a wrist wearable can now show blood pressure or blood glucose estimates without Food and Drug Administration clearance if the feature is framed for wellness and not for diabetes care or diagnosis. MedTech Dive reported that the January 6, 2026 guidance gave examples of blood pressure and blood glucose features that can qualify when they are limited to non-medical use. (medtechdive.com) Doctor-facing software sits on a different line. The Food and Drug Administration’s March 2026 town hall said clinical decision support can avoid device regulation only if it fits the statute’s exclusion criteria, including letting the health professional independently review the basis for the recommendation. (fda.gov) The January 2026 update did give one small break. In the town hall transcript, the Food and Drug Administration said it would use enforcement discretion for software that gives one clinically appropriate output if it otherwise meets the rest of the statutory criteria, which is softer than the older reading that treated a single recommendation as regulated by default. (fda.gov) That softer reading is why some headlines said Washington was easing up. STAT reported on January 6, 2026 that Commissioner Marty Makary announced changes meant to let more clinical decision support tools and wearables reach market without review, and said the agency wanted regulation to move at “Silicon Valley speed.” (statnews.com) But the same agency spent 2025 building more machinery for formal review, not less. In August 2025, the Food and Drug Administration finalized guidance on “predetermined change control plans,” which let an approved artificial intelligence device update over time only if the manufacturer lays out in advance what will change, how it will be tested, and how safety will be protected. (fda.gov) In January 2025, the Food and Drug Administration also issued draft guidance telling companies what to include in marketing submissions for artificial intelligence-enabled device software across the product life cycle. That is the opposite of a hands-off approach: it is a thicker instruction manual for products that cross into clinical territory. (fda.gov) The agency is also making the regulated market more visible. Last week, the Food and Drug Administration updated its public list of authorized artificial intelligence-enabled medical devices and said it plans future methods to identify devices that use foundation models, including large language models and multimodal systems. (fda.gov) So the message from Washington is narrower than “deregulation.” If your product counts steps, sleep, or nutrition trends, January 2026 gave you more room; if your product nudges a doctor or patient toward a diagnosis or treatment, the Food and Drug Administration still expects evidence, a submission pathway, and ongoing controls for how the model changes after launch. (fda.gov)
