FDA Has Now Authorized Over 1,451 AI-Enabled Medical Devices

In February 2026, the U.S. Food and Drug Administration’s publicly maintained AI/ML-based Software as a Medical Device (SaMD) Action Plan database crossed a significant milestone: 1,451 authorized AI-enabled medical devices. This number represents more than a decade of regulatory evolution — and a fundamental shift in how medicine is practiced at the clinical front line.

The Numbers Behind the Milestone

The FDA’s tracker, maintained by the Center for Devices and Radiological Health (CDRH), shows that AI/ML device authorizations have grown dramatically over the past five years. In 2020, the total number of authorized AI medical devices stood at approximately 400. By 2023, it had doubled. The 295 new devices authorized in 2025 alone represent the highest single-year total in the tracker’s history.

76% of all FDA-authorized AI medical devices target radiology and imaging — a concentration that reflects where AI has achieved its most validated clinical utility.

The remaining 24% spans a range of specialties: cardiology, pathology, ophthalmology, gastroenterology, and increasingly, clinical decision support tools integrated directly into electronic health record systems.

What “Authorization” Actually Means

It is important to distinguish between FDA authorization pathways. The majority of AI medical devices clear through the 510(k) pathway, which requires demonstrating substantial equivalence to an already-legally-marketed predicate device. A smaller number proceed through the De Novo pathway (for novel, lower-risk devices with no predicate) or the more rigorous Premarket Approval (PMA) process.

The 510(k) dominance has implications for AI: it means many devices are compared to earlier software tools rather than evaluated as entirely novel diagnostic systems. The FDA is actively working to address this through its Predetermined Change Control Plan (PCCP) framework, which allows manufacturers to define in advance how their AI models can be updated post-authorization.

Aidoc and the Integrated Workflow Model

Among the most clinically deployed AI companies in the FDA tracker is Aidoc, whose CARE1 platform represents a new architectural approach: rather than single-modality detection tools, CARE1 integrates multiple AI algorithms into a unified clinical workflow. It flags pulmonary embolism, intracranial hemorrhage, aortic dissection, and incidental findings simultaneously — triaging cases across specialties rather than within them.

This integration model is increasingly where the market is heading. Hospitals purchasing multiple single-purpose AI tools have encountered what researchers at Johns Hopkins have called “AI fragmentation” — siloed outputs that clinicians struggle to incorporate into unified workflow decisions.

The Radiology Concentration: Why 76%?

The disproportionate concentration of AI devices in radiology is not accidental. Medical imaging has three characteristics that make it uniquely amenable to AI development: abundant standardized training data (DICOM images), clear ground truth labeling (radiologist reports), and measurable performance benchmarks (sensitivity, specificity, AUC).

Pathology is following closely. Digital pathology platforms that analyze whole-slide images for cancer detection are now clearing FDA review with increasing frequency. Companies like PathAI and Paige.AI have demonstrated that deep learning models can detect prostate cancer and identify metastatic spread in lymph nodes with performance approaching that of experienced pathologists.

What 2026 Looks Like on the Ground

Hospital systems deploying FDA-authorized AI tools report variable integration outcomes. A 2025 implementation audit at a 12-hospital integrated delivery network found that AI tools flagging critical findings reduced time-to-treatment for pulmonary embolism by 32 minutes on average — a clinically meaningful reduction in a condition where every minute affects mortality risk.

However, the same audit found that tools deployed without corresponding changes to clinical workflow produced minimal impact. AI that surfaces a finding but does not change how that finding reaches the appropriate clinician is AI that does not change outcomes.

What the FDA Does Not Yet Measure

Authorization counts are not the same as deployment counts. The FDA tracker tells us how many devices have been cleared; it does not tell us how many are actively used, in how many facilities, on how many patients. Real-world performance monitoring remains a critical gap in the U.S. AI medical device regulatory framework.

The FDA’s Total Product Life Cycle (TPLC) advisory program and its Digital Health Center of Excellence are working toward post-market surveillance requirements for AI tools — but as of early 2026, the regulatory machinery for monitoring deployed AI performance remains underdeveloped relative to the speed of deployment.

Looking Ahead

The 1,451 figure will continue to grow. Analysts at MarketsAndMarkets project that the AI in medical imaging market alone will reach $14.7 billion by 2028, driven by continued FDA authorization, expanding reimbursement pathways, and growing clinical evidence that well-implemented AI reduces diagnostic error rates and time-to-diagnosis.

The milestone matters — but what matters more is whether those 1,451 devices are changing patient outcomes. The evidence is accumulating. The work of implementation, integration, and validation is ongoing.

Source: FDA AI/ML-based Software as a Medical Device Action Plan database, February 2026. MarketsAndMarkets AI in Medical Imaging Market Report, 2025.