Healthcare AI Hits Regulation
Healthcare AI is moving from promise to regulated use — the FDA cleared an AI ultrasound tool that estimates gestational age, even as insurers and Medicare increasingly use AI in coverage decisions and face lawsuits over denials. That combination — formal device clearances plus legal and privacy challenges around tools that record or decide care — is turning adoption into a governance and compliance problem. (diagnosticimaging.com) (kffhealthnews.org) (arstechnica.com)
A hospital can now buy one artificial intelligence tool that reads an ultrasound and estimates how far along a pregnancy is, while another artificial intelligence system at an insurer can decide whether a scan or surgery gets paid for. In 2026, those tools are no longer being treated like experiments; they are being treated like regulated products and legal risks. (fda.gov) (kff.org) The medical side of this starts with a simple problem: pregnancy care runs on timing. Doctors use gestational age, which means how many weeks a pregnancy has progressed, to decide when to order tests, give steroids, induce labor, or transfer a patient to a higher-level hospital. (butterflynetwork.com) (fda.gov) Ultrasound is the picture-taking part of that process. An artificial intelligence model can be trained on huge numbers of ultrasound images the way a face-unlock system is trained on many faces, except here it is learning patterns that match fetal development at different weeks of pregnancy. (diagnosticimaging.com) (fda.gov) On March 30, 2026, Butterfly Network said the Food and Drug Administration cleared its Gestational Age Tool, which is built into the company’s handheld ultrasound device. Butterfly said the software is the first Food and Drug Administration-cleared blind-sweep ultrasound artificial intelligence tool for estimating gestational age. (butterflynetwork.com) (diagnosticimaging.com) Blind sweep means the operator moves the probe across the abdomen in a standard motion instead of stopping to capture and measure one exact anatomy view. Butterfly said the tool can return a gestational-age estimate in under two minutes for pregnancies between 16 and 37 weeks. (appliedradiology.com) (diagnosticimaging.com) A second clearance arrived through a different regulatory path. In a February 11, 2026 letter, the Food and Drug Administration granted De Novo classification to Delivery Date AI, a prescription software device that analyzes standard pregnancy ultrasound images to produce a predicted delivery date for adults with singleton pregnancies who do not have a reliable estimated due date. (fda.gov) De Novo is the Food and Drug Administration route for a new kind of device that does not already have a close legal ancestor on the market. In that same letter, the agency put Delivery Date AI into Class II, which means it can be marketed with special controls but does not replace clinician judgment. (fda.gov) That is one half of the story: regulators are building categories for medical artificial intelligence that touches diagnosis and treatment. The other half is that payers and providers are using artificial intelligence in places where the fight is not over accuracy first, but over delay, denial, consent, and records. (fda.gov) (kffhealthnews.org) Prior authorization is the insurance checkpoint that says a doctor has to ask permission before certain tests, drugs, or procedures happen. The Kaiser Family Foundation said the Wasteful and Inappropriate Service Reduction model began on January 1, 2026 in traditional Medicare across six states and tests technologies such as artificial intelligence to review selected services through 2031. (kff.org) (kffhealthnews.org) Those six states are Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. The unusual part is that traditional Medicare had rarely used prior authorization before, while Medicare Advantage plans and private insurers had already made it routine. (kff.org) (kffhealthnews.org) The legal pressure is arriving from the clinic side too. Ars Technica reported on April 11, 2026 that several Californians sued Sutter Health and MemorialCare, alleging an artificial intelligence transcription tool recorded doctor visits without their consent and processed confidential medical conversations outside the exam room. (arstechnica.com) Put those pieces together and the new question is not whether healthcare will use artificial intelligence. The new question is which systems count as medical devices, which systems count as utilization management, which systems need patient consent, and which company is liable when a model measures, records, recommends, delays, or denies care. (fda.gov) (kff.org) (arstechnica.com)
