Lancet urges AI regulation tied to practice
The Lancet published a call to align clinical‑decision‑support AI regulation with actual clinical practice, arguing that policy needs to match how tools are used at the bedside. The recommendation focuses on closing gaps between regulatory frameworks and real‑world acute‑care workflows. (x.com)
Doctors increasingly use clinical decision support software as a bedside prompt, calculator, or warning system, and a new Lancet Digital Health paper says regulation should follow that real workflow. (thelancet.com) The paper, “Bridging the gap: aligning clinical decision support regulation with clinical practice,” was published in The Lancet Digital Health in late March 2026. Its authors wrote that ambiguous rules in the United States have led health systems to apply regulation inconsistently when these tools are used in acute care. (thelancet.com) Clinical decision support, or software that helps a clinician choose a test, flag a risk, or suggest a treatment, now sits inside electronic records and other hospital systems. A 2024 BMJ Health & Care Informatics paper said thousands of artificial intelligence models and hundreds of regulator-approved tools exist internationally, even as routine adoption remains limited. (informatics.bmj.com) The regulatory fault line is whether a tool is treated as a medical device or as software that remains outside device oversight. The Food and Drug Administration said its clinical decision support guidance explains which software functions are excluded from the legal definition of a device and which remain subject to oversight. (fda.gov) That boundary has become harder to police as hospitals test more artificial intelligence systems that do more than simple rules. The Food and Drug Administration’s public list of artificial intelligence-enabled medical devices was updated in April 2026, underscoring how quickly the regulated device market is growing. (fda.gov) The Lancet authors argue that bedside use in emergency departments, intensive care units, and other time-pressured settings does not map neatly onto static regulatory categories. They describe a gap between formal frameworks and the way clinicians actually combine software output with judgment, team discussion, and electronic record data during acute care. (thelancet.com) That argument lands as generative artificial intelligence tools move from drafting notes to offering broader clinical suggestions. A separate Lancet Digital Health commentary in 2025 said some health systems had already deployed generative artificial intelligence tools even though Microsoft said one high-profile diagnostic system was not ready for clinical use. (thelancet.com) The journal has been building this case for months. A 2024 Lancet comment on clinical decision support said low-complexity tools such as calculators and order sets are common, while more sophisticated systems are increasingly used in time-sensitive settings where fairness, safety, and accountability become harder to separate. (thelancet.com) The immediate policy backdrop is still shifting. The Food and Drug Administration said it re-issued updated final guidance on clinical decision support software on January 29, 2026, and scheduled a town hall on March 11, 2026, to discuss the changes with industry and other stakeholders. (fda.gov) The Lancet paper does not call for scrapping oversight. It calls for rules that track how clinicians actually use software at the bedside, where a prompt that looks like advice on paper can function like a decision-making tool in practice. (thelancet.com)
