AI in Medicine: Trust Gap

Artificial intelligence is already inside hospital workflows, but researchers and regulators still cannot show, at scale, that it improves patient outcomes. (technologyreview.com) In medicine, many current systems act like prediction engines: they scan images, summarize charts, draft notes, or rank patients by risk for a clinician to review. JAMA wrote in December 2025 that AI had moved rapidly into imaging, diagnostics, ambient scribing, and decision support. (jamanetwork.com) The U.S. Food and Drug Administration said in April 2026 that its public list identifies AI-enabled medical devices authorized for marketing in the United States, and a 2025 perspective in *npj Health Systems* said 1,016 such devices had been cleared as of September 2024. The same perspective said routine use still faces hesitancy centered on trust. (fda.gov) (nature.com) The evidence gap is not about whether hospitals can buy AI tools; it is about whether those tools change concrete results such as fewer complications, faster recovery, or lower mortality. MIT Technology Review reported on April 24, 2026, that adoption is outrunning proof of benefit to patients. (technologyreview.com) That trust problem gets sharper when patients do not know AI was used in their care. A JAMA perspective published in 2025 said organizations need a framework for deciding when patients should be told about AI tools, especially when the information could affect a reasonable patient’s choice. (pubmed.ncbi.nlm.nih.gov) Another concern is that people may give AI too much credit even when it is wrong. A 2025 *NEJM AI* study of 300 participants found that people could not reliably distinguish AI-generated medical responses from doctors’ responses and often rated the AI answers as more trustworthy and complete. (media.mit.edu) Researchers in *npj Health Systems* wrote in March 2025 that trust can erode if bias persists, transparency falters, or incentives misalign, and that AI can affect whether patients seek care and how they view clinicians and institutions. They called for empirical research, equitable design, and shared accountability. (nature.com) Regulators and hospital accreditors have started to answer with governance rules rather than blanket bans. The Food and Drug Administration issued draft guidance on January 7, 2025, for AI-enabled device software across the product life cycle, and JAMA said the Joint Commission and the Coalition for Health AI later called for governance committees, local validation before deployment, and continuous monitoring for drift, degraded performance, or bias. (fda.gov) (jamanetwork.com) Federal health information regulators also moved on disclosure. The Office of the National Coordinator for Health Information Technology said its HTI-1 final rule includes algorithm transparency provisions in the health information technology certification program. (healthit.gov) The Food and Drug Administration, Health Canada, and the U.K. Medicines and Healthcare products Regulatory Agency said in June 2024 that transparency for machine learning-enabled medical devices should run through the full product life cycle and should communicate information that could affect patient risk and outcomes. The agencies said that information can shape whether clinicians and patients trust a device. (fda.gov) The practical question for hospitals is no longer whether AI will appear in care. It is whether clinicians can show patients, boards, and regulators what the tool does, how it was checked on local data, and whether it is helping after deployment. (technologyreview.com) (jamanetwork.com)