What "FDA Cleared" Actually Means for an AI Tool — and How to Verify It Yourself

Every AI vendor's homepage says something about the FDA. "Cleared." "Approved." "Granted." "Registered." These are not interchangeable — they describe different levels of FDA review, and the difference decides how much weight a claim deserves. Here is the plain-language version, and the check anyone can run.

Quick answer

"FDA cleared" means the FDA found a device substantially equivalent to one already on the market, through the 510(k) pathway. "FDA approved" means premarket approval, a higher bar requiring the device's own clinical evidence. "FDA granted" or "authorized" usually means De Novo — a novel device with no predicate. "FDA registered" means only that a facility told the FDA it exists, and involves no product review at all.

The four phrases, ranked by what they prove

Approved is the strongest claim. Premarket approval (PMA) is reserved for the highest-risk devices and requires the FDA to evaluate the device's own safety and effectiveness evidence. Relatively little AI clinical software travels this road: of the 1,524 devices on the FDA's AI-Enabled Medical Device List as of July 2026, 19 carry PMA numbers.

Granted is the one most people miss. The De Novo pathway exists for novel devices with no suitable predicate — the FDA reviews the evidence directly and, if it grants the request, creates a new device type others can later use as a predicate. These carry a DEN number, and the FDA describes them as granted or authorized rather than cleared. Thirty-nine devices on that same FDA list came through De Novo. A vendor saying "FDA authorized" may be describing something real; it is simply not a 510(k) clearance.

Cleared is the common one — 1,466 of those 1,524. The 510(k) pathway asks a different question: is this device substantially equivalent to a legally marketed predicate? A clearance means the FDA answered yes and issued a substantial-equivalence determination. It authorizes marketing; it is not a finding that the device outperforms what came before, and the FDA does not call cleared devices "approved."

Registered proves the least. Establishments that manufacture devices — and importers bringing them into the country — must register with the FDA annually and list their products. Registration involves no evaluation of any product, and the FDA is explicit that registration and listing do not denote approval or endorsement. Marketing that leans on "FDA registered" as if it were a product credential is a signal to slow down.

The verification, step by step

You never have to take the phrase on faith. The records are public:

1. Ask for the submission number. A 510(k) clearance carries a K-number — the letter K, two digits for the year the FDA received the submission, then a four-digit sequence. A De Novo carries a DEN number; a premarket approval carries a P-number. A company with a real marketing authorization hands one over instantly.
2. Look it up in the matching database — the 510(k) database for K-numbers, the De Novo database for DEN numbers, the PMA database for P-numbers.
3. Open the summary — this is the step people skip. The database row itself is an index card: applicant, device name, product code, decision date, decision. The thing you actually came for — the intended use the FDA reviewed, and the predicate the equivalence argument rested on — lives in the 510(k) summary document linked from that row. Open it.
4. Compare intended use to the pitch. This is where evaluations are won or lost. Marketing language is often broader than the cleared intended use, and the gap between the two is where your risk lives.

One wrinkle worth knowing: sponsors may file a 510(k) statement instead of a summary, in which case the detailed document is not posted — the sponsor must supply it on request. If a vendor's record has no summary and they will not provide the equivalent, that itself is information.

Why this matters more with AI

AI tools compound the stakes. The marketing writes itself — "detects," "diagnoses," "catches what humans miss" — while the cleared intended use is usually far narrower, often assistance, triage, or prioritization with a clinician making the call. Reading the actual authorization record is among the highest-value habits an evaluator can build, and it costs a few minutes.

If you want to see what these categories look like in aggregate rather than one product at a time, we counted them across the public record.

We read the FDA record weekly across the AI-in-medicine categories and publish what it says, free. If a vendor claim you have been shown doesn't square with its record, tell us — those make the best posts.

Common questions