FDA AI Medical Device Approvals 2026: List and Statistics

Why This Page Does Not Publish a Fixed 2026 Device Count

FDA's list of AI/ML-enabled medical devices is updated on a rolling basis rather than published as a fixed annual report. Any specific 2026 approval count, device name list, or category percentage cited without pulling directly from that live source at the time of use is likely to be inaccurate by the time it is read. For compliance leaders, the operational implication is straightforward: treat FDA's device list as a live system of record to be queried at deployment time, not a static reference to be cached in policy documents. Internal governance processes should build in a step that re-checks a device's current authorization status, including any PCCP scope, before each new deployment or major version update, rather than relying on a point-in-time snapshot.

Regulatory Pathways: What Each One Signals

FDA authorizes AI-enabled devices through three primary premarket pathways. The 510(k) pathway requires a manufacturer to demonstrate substantial equivalence to an already-cleared predicate device, and it has historically been the most common route for AI/ML devices. De Novo authorization applies to novel, lower-to-moderate risk devices with no suitable predicate, establishing a new device classification in the process. Premarket Approval (PMA) is reserved for higher-risk devices and requires the most extensive clinical evidence. These pathways are not interchangeable in terms of evidentiary depth: a 510(k) clearance reflects equivalence to prior technology, while a PMA reflects a full clinical evidence review. Compliance teams should record which pathway underlies each deployed device, since the pathway itself is a proxy for the depth of premarket scrutiny the device has undergone.

Predetermined Change Control Plans and Ongoing Model Updates

A distinguishing feature of AI-enabled devices, compared to traditional medical devices, is that their underlying models can change after authorization. FDA's Predetermined Change Control Plan (PCCP) framework allows a manufacturer to define, in advance, a bounded scope of future model modifications, along with the verification methods and impact assessments tied to those changes, so that qualifying updates do not require a new premarket submission each time. This mechanism is significant for governance because it shifts part of the compliance burden from FDA's premarket review to the manufacturer's and deployer's ongoing monitoring obligations. If a device operates under an accepted PCCP, an enterprise deploying it should know the defined change boundaries and have a way to confirm that observed model updates fall within that pre-approved scope rather than exceeding it.

Historical Patterns as a Baseline for Evaluating 2026 Data

FDA's own device list categorization from prior years shows two consistent patterns worth using as a baseline when reviewing current data. First, the majority of authorized AI/ML devices have used the 510(k) pathway, with De Novo and PMA authorizations occurring less frequently. Second, radiology and cardiology have represented the largest clinical specialty categories among authorized devices, reflecting the maturity of imaging-based AI applications relative to other specialties. These patterns describe the shape of the historical dataset, not a confirmed 2026 breakdown, and compliance leaders should verify whether the current year's data continues to follow this distribution or diverges from it, since a shift toward De Novo or PMA authorizations, for example, would suggest FDA is treating newer device categories as higher risk or more novel.

Governance Considerations Once a Device Is Deployed

Once an FDA-authorized AI device enters a clinical or operational environment, governance responsibility does not end at the procurement decision. Compliance leaders should maintain a mapping between each device's authorization pathway and its internal risk classification, since a 510(k) clearance and a PMA approval do not carry the same evidentiary weight and should not be governed identically. Where a device operates under a PCCP, governance processes should include monitoring for model drift or updates that could exceed the pre-approved change boundaries, since an update that falls outside that scope would require a fresh submission FDA may not yet have reviewed. Audit trails for device outputs should be retained in a manner consistent with the device's labeling and any post-market surveillance conditions, since these records may be the primary evidence available if a device's behavior is later questioned. These are the same runtime governance concerns (agent or model identity, permission scope, and audit logging) that apply broadly to any AI system operating inside regulated infrastructure, and they extend naturally to FDA-authorized devices once those devices are integrated into broader clinical or enterprise workflows.