HHS Proposes Eliminating Transparency Rules for Clinical AI

The Department of Health and Human Services is moving to dismantle the first federal transparency framework for artificial intelligence used in clinical decision-making, a regulatory retreat that would leave hospitals without standardized tools to evaluate the algorithms now embedded in patient care.

In January 2026, HHS released a proposed rule—HTI-5—that would eliminate disclosure requirements finalized just two years earlier. Those requirements, part of the certification criteria for electronic health record systems, mandated that AI developers publish 31 standardized attributes describing how their tools were built, tested, and maintained. The disclosures covered training data composition, fairness testing, external validation, performance metrics, and update schedules.

The timing is notable. By the end of 2025, FDA had authorized more than 1,450 AI-enabled medical devices, with 295 cleared in 2025 alone. Two-thirds of clinicians now use AI in their work, according to the American Medical Association. Yet fewer than 2 percent of cleared devices were supported by randomized clinical trials, and studies estimate hallucination rates for large language models used in clinical decision support range from 8 to 20 percent.

Why it matters

The proposed rollback eliminates the only federal transparency mandate covering clinical AI inside certified electronic health records—the systems used by more than 96 percent of U.S. hospitals. Without standardized disclosures, hospital procurement officers and compliance teams lose a common framework for comparing vendors and auditing deployed models. The burden shifts from developers to individual health systems, each forced to construct separate due diligence processes for every algorithm in every workflow.

A governance gap widens

The transparency requirements were designed to address institutional needs, not just point-of-care information. Hospital quality committees, procurement officers, and compliance leads needed standardized data to meet their own governance obligations. The disclosure regime provided that baseline.

FDA's device authorization process, while broad in scope, was never designed to monitor AI performance after market entry. A 2025 analysis found that most 510(k) clearance summaries lack details on study design, sample sizes, and demographic representation. The source-attribute disclosures helped fill this post-market accountability gap by requiring developers to publish validation and update schedules.

Meanwhile, much clinical AI now entering hospitals falls outside FDA jurisdiction entirely. Non-device clinical decision support tools, internally developed models, and generative AI assistants frequently operate in a regulatory blind spot. The Decision Support Interventions criterion covered both FDA-regulated software and non-device predictive tools. Eliminating it leaves the fastest-growing category of clinical AI—generative tools built on foundation models—without oversight at precisely the moment hallucination rates remain in double digits.

The alternative path

A more defensible approach would preserve the disclosure requirements and integrate them into continuous monitoring infrastructure. The nine core disclosure categories map directly onto the AI risk management framework published by the National Institute of Standards and Technology and align with FDA's own post-deployment surveillance questions.

Rather than retiring these requirements, regulators could mandate that disclosures be machine-readable, versioned, and accessible through standard APIs. This would enable hospital governance committees, payers, and accreditors to verify compliance continuously, placing the marginal cost on developers who control the models.

The comment period on HTI-5 has closed, but the rule is not yet final. The past three years of deployment have demonstrated that clinical AI may not fail at regulatory clearance but can fail quietly over months as populations shift and models drift.

These details were first reported by The Regulatory Review.