FDA-Cleared AI Medical Devices: What They Actually Do

This content is educational. It describes the regulatory landscape for AI in medicine. It is not medical advice and is not a guide to using any specific device.

Why this matters

Consumer health AI sits on a regulatory spectrum from "not regulated" to "fully regulated medical device." Most of the AI tools consumers encounter — symptom checkers, fitness coaches, wellness chatbots — sit on the unregulated end. A smaller but rapidly growing set of AI tools have crossed into regulated medical-device territory and carry FDA clearance.

The distinction matters because it tells consumers and clinicians what testing the technology has been through, what claims it's authorised to make, and what oversight applies to its ongoing use. A device with FDA clearance has met a regulatory standard. A wellness app calling itself "AI-powered" has not. Both can be useful; the assurance level differs dramatically.

This page describes how AI medical device clearance actually works, what kinds of AI have been cleared so far, and how to read the difference between regulated and unregulated AI health tools.

The basic FDA framework

In the US, the Food and Drug Administration regulates products that meet the legal definition of a medical device. Software that meets this definition is regulated as Software as a Medical Device (SaMD).

The 1976 Medical Device Amendments and subsequent statutes define a medical device by its intended use:

• Diagnosis of disease or other condition
• Cure, mitigation, treatment, or prevention of disease
• Affecting the structure or function of the body

Software intended for any of these purposes can be a medical device. Software for general wellness, fitness, or information falls outside this framework.

The 2016 21st Century Cures Act explicitly carved out certain low-risk software functions from device regulation — including general wellness software that doesn't make disease claims [FDA 2019]. This is why most consumer health apps avoid disease language: it keeps them on the unregulated side.

The clearance pathways

For software determined to be a medical device, the FDA has several clearance pathways:

510(k) clearance

The most common pathway. The manufacturer demonstrates that the device is "substantially equivalent" to a previously cleared device (a "predicate"). The standard is similarity to existing devices, not de novo demonstration of safety and effectiveness.

Most AI medical devices cleared to date have come through 510(k). The benchmarks are typically:

• The algorithm performs comparably to the predicate on standard test datasets
• The intended use is similar
• Safety considerations are addressed

The pathway is faster and less expensive than the alternatives. Critics have argued that 510(k) clearance for novel AI may be inadequate when the predicate is much older and less capable [Wu et al. 2021].

De Novo classification

For novel devices with no predicate but low-to-moderate risk. The manufacturer must demonstrate safety and effectiveness directly. Slower and more expensive than 510(k) but provides stronger validation.

Several pioneering AI devices have used this pathway, including IDx-DR (the first autonomous diagnostic AI cleared by the FDA in 2018).

Premarket Approval (PMA)

The most rigorous pathway, required for high-risk Class III devices. Manufacturer must demonstrate safety and effectiveness through clinical trials. Slow and expensive. Few AI devices have come through this pathway because most AI applications have been classified as Class II.

Breakthrough Device designation

A relatively new program that provides expedited review for devices addressing serious conditions where existing alternatives are inadequate. Several AI devices have received this designation, though it doesn't lower the substantive evidentiary requirements.

What's actually been cleared

As of 2024, the FDA has cleared more than 950 AI-enabled medical devices [FDA AI/ML Database 2024]. The distribution by category:

Radiology dominance reflects two things: imaging produces structured data well-suited to AI, and radiology has long had quantitative measurement tools that AI can extend. Most cleared AI devices augment radiologist workflow — flagging potential abnormalities for review, measuring specific structures, or prioritising urgent cases.

Examples of cleared AI devices

IDx-DR (now LumineticsCore) — autonomous detection of diabetic retinopathy from retinal images. First fully autonomous AI diagnostic cleared by the FDA (April 2018). Used in primary care to screen patients without ophthalmologist involvement.

Viz.ai — stroke triage system. Analyses CT angiograms for large vessel occlusion strokes and notifies the stroke team. Multiple clearances for related indications.

Apple Watch ECG — single-lead ECG with atrial fibrillation detection. Cleared for irregular rhythm notification (Class II).

Caption AI (Caption Health, now part of GE) — ultrasound image acquisition guidance for non-specialist users. Allows nurses and primary care providers to capture cardiac ultrasound images of diagnostic quality.

Tempus xT — AI-assisted molecular tumour profiling for cancer treatment selection.

Aidoc — multiple AI clearances for radiology workflow prioritisation across different conditions (intracranial haemorrhage, pulmonary embolism, etc.).

These are examples, not exhaustive. The FDA maintains a public database of cleared AI/ML-enabled devices that is updated regularly.

What clearance actually guarantees

A cleared AI medical device has demonstrated:

• The intended use is appropriate for medical device regulation
• The algorithm meets defined performance criteria on validation datasets
• Manufacturer quality systems meet FDA standards
• Risk management documentation is in place
• Labelling and intended use are appropriate

What clearance does not guarantee:

• That the AI will perform equally well on populations not represented in validation data
• That clinical implementation will produce the same results as the validation study
• That the device is "better than" alternatives
• That long-term performance will match initial validation
• That the AI is appropriate for any specific patient

The clearance establishes a regulatory floor. Real-world performance, equity considerations, and ongoing monitoring remain active areas of research and oversight [Wu et al. 2021; Adamson & Smith 2018].

The continuous-learning challenge

Traditional medical devices have fixed designs that don't change once cleared. AI presents a regulatory challenge because the technology can be designed to continuously learn from new data — improving (or degrading) over time without explicit modification.

The FDA has been developing a framework for handling this. The current approach:

• AI devices cleared with "locked" algorithms (fixed at clearance) follow standard rules
• For devices designed to continuously learn, the FDA's "Predetermined Change Control Plan" (PCCP) framework, finalised in late 2024, allows manufacturers to specify in advance how the algorithm will be updated post-clearance, with associated monitoring requirements [FDA 2024 PCCP guidance]

Most cleared AI medical devices today still use locked algorithms. The PCCP framework is being adopted gradually.

How EU and UK regulation compares

The EU Medical Devices Regulation (MDR) 2017/745 captures AI as a medical device using similar definitional principles. The MDR requires:

• CE marking before market entry
• Notified body review for higher-risk classes
• Clinical evaluation evidence
• Post-market surveillance

The EU has historically taken a different approach to AI medical devices: similar regulatory rigor but with the additional EU AI Act overlay (in force from 2024-2026 in phases) which classifies AI in healthcare as a "high-risk" AI system with additional transparency, oversight, and accountability requirements.

The UK MHRA inherited the EU framework post-Brexit and has been developing its own AI medical device guidance.

The pattern: substantially similar regulatory rigour across major Western jurisdictions, with implementation details varying. Consumer-facing devices selling globally generally need separate approvals in each major market.

How to read AI health tools as a consumer

Practical signals for evaluating AI health tools:

1. FDA clearance status — searchable in the FDA's 510(k) database and AI/ML device list. Cleared status is a meaningful threshold.
2. CE marking — for EU/UK markets, equivalent regulatory threshold.
3. Published validation studies — peer-reviewed evidence is a strong signal regardless of regulatory status.
4. Intended use language — clearance is specific to a defined indication. Off-label use is not what the clearance covers.
5. Population studied — clearance validation cohorts have specific demographic and clinical profiles. Performance can degrade outside those profiles.
6. Wellness vs medical positioning — apps making explicit medical claims should have regulatory status. Apps positioned as "wellness" or "information only" typically don't.

For consumers: cleared AI devices have crossed a meaningful threshold. Wellness AI apps may still be useful for what they're designed to do — but the assurance level is different and shouldn't be conflated.

What Proco's position in this landscape is

Proco Scanner is positioned as an information tool — describing what published research says about supplement ingredients. It does not diagnose, treat, recommend treatment, or claim therapeutic effects. This positioning is deliberate and keeps Proco Scanner outside the medical device classification.

This is a choice. The Scanner could theoretically be developed toward indicated medical-device use (e.g., "screening for ingredients contraindicated with prescription medications") — but that would require a full clearance pathway, clinical validation, ongoing regulatory oversight, and a substantially different product. For v1.0, the information positioning matches the actual product capability.

For readers evaluating the AI health space more broadly: the regulated and unregulated tools are solving different problems with different evidence levels. Confusing them — assuming "AI health tool" means equivalent oversight — is a common consumer error.

Related Proco pages

• The state of consumer health AI in 2026
• How supplements are regulated: EU vs US vs UK
• How to read a clinical trial

Sources

Proco provides educational, research-based information. This page describes the regulatory landscape for AI medical devices. It is not medical advice or guidance for use of any specific device. If you encounter an AI health tool and want to understand its evidence base, look for the regulatory clearance status and published validation literature.

Schema (for implementation)

{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "FDA-Cleared AI Medical Devices: What They Actually Do",
  "description": "More than 950 AI-enabled medical devices have FDA clearance, mostly in radiology. This page describes the clearance pathways, what clearance guarantees, and how to read the difference between regulated AI and wellness apps.",
  "datePublished": "2026-06-01",
  "dateModified": "2026-05-31",
  "author": {"@type": "Organization", "name": "Proco"},
  "publisher": {"@type": "Organization", "name": "Proco", "url": "https://procohq.com"},
  "about": {"@type": "Thing", "name": "FDA-cleared AI medical devices"}
}