Whether a healthcare AI product needs FDA clearance depends on its intended use, not on the fact that it uses AI. If the software is intended to diagnose, treat, cure, mitigate or prevent disease, or to drive a clinical decision, it is likely a medical device and may require FDA clearance or approval. If it supports administrative work, wellness, or provides information that a clinician independently reviews, it may fall outside FDA device regulation. The deciding factor is what the product claims to do and how much a clinician relies on it.
This matters because getting it wrong is costly in both directions: shipping a device-like AI without clearance is a serious regulatory risk, while treating an administrative tool as a device wastes time and money. Importantly, this guide is educational, not regulatory or legal advice. It explains how the FDA generally thinks about AI so you can have an informed conversation with qualified regulatory experts, who make the actual determination.
How does the FDA decide if AI is a medical device?
The FDA looks primarily at intended use: what the product is designed and marketed to do. Software that is intended to diagnose, treat or inform a clinical decision about a disease or condition generally meets the definition of a medical device, including Software as a Medical Device (SaMD). Software that does not make such a claim generally does not.
Two related questions shape the answer. First, what does the product claim, in its labelling and marketing, that it does? Second, how much does a clinician rely on its output, does the clinician independently review the basis for a recommendation, or act on it directly? Higher clinical reliance and stronger clinical claims push a product toward being a regulated device. This is why the same underlying model can be a device in one product and not in another.
Which AI products usually do not need FDA clearance?
AI products that support administrative work, wellness, or that simply inform a clinician who independently reviews the reasoning often fall outside FDA device regulation. These are lower-risk uses where the software is not making or driving a clinical decision.
Common examples that are frequently not devices (subject to their specific claims and design):
- Administrative automation. Scheduling, prior authorization, claims and billing tools.
- Ambient documentation. AI that drafts notes a clinician reviews and approves.
- General wellness. Tools for general fitness or wellbeing without disease claims.
- Certain clinical decision support. CDS that displays information a clinician can independently review, and does not drive time-critical decisions, may qualify as non-device under US criteria.
The caveat matters: "usually not a device" depends entirely on the specific claims and how the tool is used. A documentation tool that started giving diagnostic suggestions could cross the line. Design and marketing decisions change the answer.
Which AI products usually do?
AI products that diagnose, detect disease, drive treatment, or that a clinician relies on without independently reviewing the reasoning generally are medical devices and typically need FDA clearance or approval. These are higher-risk, clinically consequential uses.
Common examples that are frequently devices (again, subject to specifics):
- Diagnostic AI. Software that detects or diagnoses a condition, for example from an image.
- Triage that drives urgent action. AI that prioritises patients for time-critical care.
- Treatment-driving software. Tools that recommend or adjust treatment relied upon directly.
- Continuous monitoring with clinical alerts. Software that detects deterioration and drives clinical response.
If your AI makes a clinical claim or a clinician acts on its output without independently checking the reasoning, assume it may be a device and get expert input early.
What are the FDA pathways for AI medical devices?
If your AI is a medical device, the route to market depends on its risk class. There are three main pathways.
| Pathway | For | In short |
|---|---|---|
| 510(k) | Moderate-risk devices with a predicate | Show substantial equivalence to an existing device |
| De Novo | Novel low-to-moderate-risk devices | Establish a new device type and controls |
| PMA (Premarket Approval) | High-risk devices | The most rigorous route, strong clinical evidence |
The right pathway is a regulatory determination based on the device's risk and novelty. Most AI-enabled devices to date have come through the 510(k) or De Novo routes. Your regulatory experts decide the classification and pathway; we build the software and evidence to support it.
Intended use decides whether AI is a device, qualified regulatory experts confirm it, and building to the right quality bar from the start is what keeps you safe. As an engineering partner we build and evidence; we do not provide regulatory or legal advice.