The FDA keeps a public list of AI medical devices. It has 1,610 entries

The "AI-Enabled Medical Devices" list is not a law. The FDA runs it as a registry, a public index of devices with an artificial intelligence component that have been cleared for marketing. In mid-September 2026 the page carried the date "content current as of 09/04/2026," and the table counted 1,610 entries. The oldest entry is dated 15 March 2001. The newest reach the beginning of September 2026. In 2026 alone the agency added 182 decisions.

The list runs from the most recent decisions. At the top, dated 29 June 2026, sit Auto-Seg (SO-0012), Spine from Agada Medical (number K253628, radiology panel), ADAS 3D from Adas3D Medical (K253969) and the Alturion Series Diagnostic Ultrasound System from Philips Ultrasound (K260667). A day later the agency cleared SOMNUM from Honeynaps (K253390, neurology panel). On 26 June came MammoScreen (5) from Therapixel (K260714), a mammography tool.

Further down the table are syngo.CT Coronary Cockpit from Siemens Medical Solutions (K253786, 25 June), Enhanced Boundary for PCCT from GE Medical Systems (K253560, 24 June) and CMN Capillary Function with Oncology Virtual Expert from Cercare Medical (K260316, 22 June). The 18 June decisions include RadOncAI (K253050), ARTHUR (K260190) and Vascular Assist Occlusion Triage (K260729). The registry covers diagnostic models, then, but also software that supports treatment planning and triage.

What is not on the list yet

The FDA says a future update will flag devices that use functions built on large language models. Until then, manufacturers are encouraged to put the relevant information in their public summaries. An entry on the list does not amount to full documentation either. The agency notes that cleared devices whose decision summary was not published during the data collection period only appear at the next update.

Advisory committees are a separate track. On 23 September 2026 an FDA advisory panel backed the Galleri blood test from Grail, which screens for multiple cancer types. The panel voted 10 to 0 that there is "reasonable assurance" of safety, 6 to 4 that there is reasonable assurance of effectiveness, and 7 to 2 with one abstention that the benefits outweigh the risks in patients who meet the criteria of the proposed indication. The agency does not have to follow a panel's opinion, but it often does.

In clinical practice, what counts is the scope of the evidence, not the number of entries. The registry says how many devices were cleared. It does not say whether each of them improves treatment outcomes. The public list is a starting point, not the end of the assessment.