Critical Results Shouldn't Sit in a Fax Queue

Intelligent clinical diagnostics refers to AI and machine learning software that helps clinicians detect, classify, or prioritize a medical condition — most commonly by analyzing medical images, lab results, or continuous patient data and flagging findings for a clinician's review. It spans a wide range of maturity, from decision-support tools that surface a possibility for a human to evaluate, to more autonomous triage systems that reorder a worklist based on urgency.

Yes — AI-enabled diagnostic software is regulated by the FDA as Software as a Medical Device (SaMD), generally requiring 510(k) clearance, De Novo authorization, or Premarket Approval before it can be marketed. The FDA's AI-Enabled Medical Device List had grown to roughly 1,451 cumulative authorizations by the end of 2025, with about 97% cleared through the 510(k) pathway — making it one of the fastest-growing categories in the FDA's device portfolio.

The evidence is more mixed than marketing claims often suggest: a 2026 meta-analysis of randomized controlled trials found AI-based clinical decision support produced only a modest, statistically marginal improvement in diagnostic accuracy overall, with meaningful benefit concentrated in specific use cases like chest radiology and largely negligible effects in emergency and general medicine settings. AI diagnostic performance varies significantly by clinical use case and system design — broad, category-wide accuracy claims should be treated with caution.

Under current law, physicians and hospitals carry most of the liability exposure — physicians remain primarily responsible for the clinical judgment exercised even when using an AI tool, and hospitals face risk if they fail to properly validate, train staff on, or govern the systems they deploy. AI vendors face liability mainly through product-liability claims involving design defects or inadequate warnings. Legal experts generally agree the law is still evolving faster than clear precedent has been set.

Reimbursement exists but remains inconsistent: as of January 2026, only 3 of 26 clinical AI CPT codes hold permanent Category I status with guaranteed payment, while the rest are Category III temporary codes. CMS has proposed a new Software as a Medical Service (SaMS) payment category for 2027 to standardize AI reimbursement. Some tools — including a continuous AI sepsis monitor — have received Medicare New Technology Add-on Payment (NTAP) eligibility for inpatient use.

For the large majority of currently deployed tools, yes — both FDA clearance pathways and current medical liability standards are built around a clinician reviewing and confirming AI-flagged findings rather than acting on them automatically. Fully autonomous diagnostic AI exists only in narrow, specifically authorized use cases. A 2025 VA OIG review specifically flagged limited oversight of AI performance after deployment as a systemic gap, reinforcing that human review and ongoing monitoring remain essential safeguards.