FDA clearances outpace proof

The Food and Drug Administration cleared 295 artificial intelligence-enabled medical devices in 2025, but published evidence still shows many cleared tools reached the market with limited clinical testing. (fda.gov; intuitionlabs.ai) Artificial intelligence medical devices are software or hardware tools that turn patient data into predictions, alerts, or measurements, such as image triage for stroke scans or sensors that estimate heart attack risk. The Food and Drug Administration reviews them through device pathways such as 510(k) clearance, De Novo classification, or premarket approval. (fda.gov) A 2025 JAMA Network Open study examined 903 Food and Drug Administration-authorized artificial intelligence devices listed through August 31, 2024 and found clinical performance studies were reported for 505 devices, or 55.9%. Another 218 devices, or 24.1%, explicitly said no performance studies had been conducted. (jamanetwork.com) The same study found radiology dominated the field, with 692 of 903 devices, or 76.6%, followed by cardiovascular medicine with 91 and neurology with 29. Less than one-third of the clinical evaluations reported sex-specific data, and about one-fourth reported age-related subgroup data. (jamanetwork.com) That gap sits inside a market that is still expanding fast. By July 2025, the Food and Drug Administration’s public list included 1,247 approved artificial intelligence-enabled technologies, and Food and Drug Administration official Robert Ochs said more than 75% were tied to radiology. (rsna.org) The agency updated its artificial intelligence device framework on January 7, 2025 with draft guidance that tells developers to manage risk across the total product life cycle, from design and marketing submission to postmarket monitoring. The guidance is draft and nonbinding, but it lays out what information the agency wants in submissions for artificial intelligence-enabled device software functions. (fda.gov) The pressure point is the 510(k) pathway, which clears many moderate-risk devices by comparing them with an existing product rather than requiring new prospective human trials. A 2025 JAMA Health Forum study said nearly all cleared artificial intelligence-enabled medical devices entered through 510(k), and noted that the pathway does not require prospective human testing. (jamanetwork.com) That study looked at 950 artificial intelligence medical devices authorized through November 2024 and found 60 devices were linked to 182 recall events. About 43% of those recalls happened within one year of authorization, and diagnostic or measurement errors were the most common cause. (healthcaredive.com; jamanetwork.com) Researchers behind the recall study said devices without clinical validation were more likely to be recalled than tools backed by retrospective or prospective studies. They also found public companies accounted for about 53% of devices on the market but more than 90% of recall events in their sample. (healthcaredive.com) The Food and Drug Administration says its public list is meant to improve transparency, but it also warns the list is not comprehensive and that public summaries do not include most of what sponsors submit. For hospitals deciding whether to buy these tools, that leaves clearance as a regulatory threshold, not a full public record of how a product performs in routine care. (fda.gov)