FDA Cleared, FDA Approved, FDA Authorized

A field guide to healthcare AI’s most misused words, and the evidentiary realities they obscure.

Presented By Our Partners

In April of 2018, an artificial intelligence system became the first autonomous diagnostic in any field of medicine to receive FDA authorization. The system, developed by a small company in Coralville, Iowa, was designed to read retinal images and diagnose diabetic retinopathy without any human oversight at all. A primary care physician, or a nurse, or a trained operator at a retail clinic could feed images from a fundus camera into the algorithm and receive, within seconds, a diagnostic decision. The authorization was the product of years of development, venture funding, and a preregistered prospective clinical trial of nine hundred patients across ten primary care sites, with results compared against the Wisconsin Fundus Photograph Reading Center as the reference standard. The trial was published in npj Digital Medicine later that year. The FDA pathway, called De Novo, is reserved for novel devices that have no existing predicate, and the agency typically requires the kind of prospective evidence that would also satisfy a traditional clinical trial reviewer.

In the years that followed, hundreds of other AI products received what is also described as FDA authorization. Most of them came through a different pathway, the 510(k), which requires the applicant to demonstrate that the new device is substantially equivalent to an existing legally marketed device. The 510(k) pathway does not require a prospective clinical trial. It does not, in most cases, require any human subjects research. It requires analytical validation, typically on retrospective data, sufficient to show that the new device’s outputs are similar to those of a predicate that may itself have been cleared on substantial equivalence to a still earlier predicate.

Both kinds of products can accurately describe themselves as FDA authorized, but only the 510(k) product is FDA cleared in the strict regulatory sense. The broader phrase is technically correct in both cases. It is also, in both cases, used as though it conveyed something specific about the strength of the clinical evidence underneath. In one of the two cases, it often does. In the other, it often does not. The consumer reading the homepage cannot tell which is which.

This piece is a working glossary. Six terms appear regularly in healthcare AI marketing, longevity product copy, and the press releases that fill in between. The terms are not interchangeable. They are not equally rigorous. They are not, in most cases, even adjacent to one another on the evidentiary spectrum. Confusing them is the single most common factual mistake the public and the press make about healthcare regulation, and the confusion is, by design, the foundation on which a significant amount of marketing depends.

The six terms, in order of seriousness

FDA Approved

The phrase FDA approved has a specific technical meaning. It applies to prescription drugs that have completed a New Drug Application or Biologics License Application, and to a small subset of medical devices, generally the highest risk Class III devices, that have completed a Premarket Approval, or PMA, submission. These pathways require substantial evidence of safety and effectiveness before commercial sale is permitted, often including prospective clinical studies with meaningful sample sizes.

In the universe of healthcare AI products, FDA approved applies to almost none of them. Of the more than one thousand AI and machine learning enabled medical devices the FDA had authorized by early 2025, fewer than one percent came through the PMA pathway. Most healthcare AI products are not Class III devices. They do not, under current rules, need to clear the PMA bar. The phrase FDA approved, when it appears in the marketing copy of an AI product that has not been through a PMA, is either a misuse or a deliberate sleight of hand. The reader who notices the misuse has caught something specific.

A few additional cases are worth flagging. Some AI products are downstream components of an approved drug or device, and the parent product’s approval status is sometimes invoked to imply that the AI component itself is approved. This is a category error. The approval attaches to the parent product. Some marketing materials cite FDA approval to mean FDA accepted our IDE for a clinical trial, where an IDE is an Investigational Device Exemption, a permission to study an unapproved device in humans. The IDE is not an approval. It is a permission to gather the evidence that may, eventually, support one. A reader who encounters FDA approved in healthcare AI marketing should default to skepticism and look for a PMA number, a drug application number, or other specific regulatory citation. Absent those, the phrase is almost certainly wrong.

FDA Cleared

The phrase FDA cleared almost always means 510(k) clearance, the pathway codified in 1976 under the Medical Device Amendments to the Food, Drug, and Cosmetic Act. The 510(k) standard is substantial equivalence. The applicant identifies a predicate device, an existing legally marketed product, and demonstrates that the new device is similar enough to the predicate in intended use, technological characteristics, and performance to be regulated under the same controls. The clearance does not certify safety in the sense the lay reader imagines. It certifies that the new device is not substantially different from a device the FDA has already permitted on the market.

This pathway accounts for the vast majority of FDA authorized healthcare AI. Multiple landscape analyses, including a comprehensive 2024 study of the ML enabled Class II devices the FDA authorized that year, have found that between ninety four and ninety seven percent of healthcare AI authorizations have come through 510(k). The same 2024 analysis found that of one hundred sixty eight authorized devices, only twenty nine percent reported both sensitivity and specificity in their public summary documents, and only sixteen percent provided demographic data on the validation populations.

A subtle and important point about predicates: the predicate device a new applicant cites may itself have been cleared on substantial equivalence to an earlier predicate, in a chain that can stretch back decades. The chain does not, in most cases, terminate at a device with prospective clinical evidence. It terminates at a device that was on the market before 1976 or that was cleared via De Novo or PMA at some point in the past. A 2025 analysis of the 2024 ML enabled device cohort found a median predicate age of just over two years, with the predicate chain commonly involving devices whose own original clearances had been issued years or decades earlier. The reader who internalizes that 510(k) means substantially equivalent to something that was substantially equivalent to something has understood the pathway correctly.

None of this makes 510(k) clearance worthless. The pathway is appropriate for genuinely incremental devices and for product categories where the underlying risk profile is well understood. It is, however, importantly different from PMA and from De Novo, and the marketing language that treats cleared as a synonym for approved is exploiting a confusion the FDA has tried, with mixed success, to discourage. The agency itself, in published guidance, has noted that clearance and approval are not interchangeable terms.

FDA Authorized

The phrase FDA authorized is an umbrella. In healthcare AI, it can refer to 510(k) clearance, De Novo authorization, PMA approval, or, in emergency contexts, Emergency Use Authorization. The word is useful because it describes products the FDA has allowed onto the market without collapsing the very different pathways by which that happened. The word is also dangerous because marketing copy can use it to avoid specifying the pathway at all.

The De Novo pathway is the most consequential non-510(k) pathway for many novel healthcare AI devices. It exists for novel low to moderate risk devices that lack a suitable predicate. Because there is no predicate, substantial equivalence is not available, and the applicant must instead submit evidence sufficient to support special controls that the FDA will then issue to govern the device category. In practice, De Novo authorization often requires prospective clinical evidence at a level closer to PMA than to 510(k), though the formal evidentiary standard is lower. The IDx-DR pivotal trial described at the top of this piece is the canonical example. The De Novo authorization the FDA granted in April of 2018 became the basis on which subsequent diabetic retinopathy AI products could file 510(k) applications, citing IDx-DR as the predicate. The chain begins, in this case, with a rigorous prospective trial. In other product categories, the chain begins differently.

Emergency Use Authorization, the other major form of authorization, is even more confusing in the marketing copy of products that hold one. An EUA is not a clearance and not an approval. It is a temporary permission, issued under a public health emergency, that allows a product onto the market on the basis of available evidence that may be substantially less than would be required for standard clearance or approval. Many EUAs from the COVID pandemic have been revoked. Many were converted, after additional evidence, into standard clearances. The phrase FDA authorized, in a product that holds only an EUA, is technically correct but informationally incomplete.

FDA Granted Breakthrough Device Designation

This is the most consistently misused phrase in healthcare AI marketing. Breakthrough Device Designation is an expedited review program. It is granted to devices that may offer significant advantages over existing approved or cleared alternatives in the treatment or diagnosis of life threatening or irreversibly debilitating conditions. The designation entitles the applicant to additional FDA interaction during product development. It does not, in itself, mean the device has been cleared or approved. It is closer to a calendar than a verdict.

A company that has received Breakthrough Designation will often say, in marketing materials and press releases, that the FDA has granted the device some special status. The grant is real. The status is real. The device is not, on the basis of the designation alone, legally cleared or approved for commercial sale. A reader who encounters FDA Granted or FDA Designated in a healthcare AI press release should default to assuming that the phrase refers to Breakthrough Designation, which is meaningful but distinct from the clearance or approval the consumer is likely to assume. Some Breakthrough Designated devices have gone on to receive full clearance or approval. Many have not yet done so.

Featured Partner

Invest in the Infrastructure Behind Modern Medicine

As healthcare expands beyond hospital walls, the buildings and campuses supporting that shift are generating compelling returns for investors who move early. The Healthcare Real Estate Fund offers qualified investors direct access to a curated portfolio of medical office, outpatient, and specialty care facilities.

Learn More →

FDA Registered / FDA Listed

This is the phrase the supplement industry and a significant fraction of consumer health products use when no real regulatory pathway applies. FDA registered and FDA listed mean that the manufacturer or facility has submitted basic information to the agency. The submission is mandatory for facilities that handle FDA regulated products. It does not involve any review of the product itself. It does not involve any testing. It does not involve any clearance or approval. It is paperwork.

The FDA has issued public communications repeatedly warning about misleading use of FDA registered in marketing. In March of 2021, the agency formally called on certain firms to stop producing and issuing what it called misleading FDA Registration Certificates, which had been used to suggest a level of regulatory endorsement that does not exist. The fact that the agency had to do so suggests how widespread the practice had become.

For the careful reader, FDA registered should function as a negative signal. A product that uses this phrase in its marketing has typically chosen it precisely because the stronger phrases, FDA cleared or FDA approved, do not apply. The choice is deliberate. The implication is misleading. Most consumers do not parse the difference.

Not Evaluated by the FDA

The phrase the FDA itself requires on dietary supplement labels reads as follows: These statements have not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease. The phrase exists because the Dietary Supplement Health and Education Act of 1994 created a regulatory category in which supplements are treated more like food than like drugs. The FDA does not review the safety or efficacy of dietary supplements before they reach the market. The agency can act after the fact, if a product is found to be unsafe or misbranded, but the structural posture is reactive rather than prospective.

A meaningful fraction of the longevity science marketplace operates within this category. The phrase clinically validated, applied to a longevity supplement, almost never means anything resembling FDA review. It typically means that a small study has been run, often by or for the company that sells the product, and that the results were positive enough to use in marketing copy. The disclaimer, in small print at the bottom of the product page, is the regulatory honest accounting. The marketing copy, with its references to clinical research and proprietary blends and patented formulations, is the rhetorical accounting. The two are rarely brought into contact.

Why the confusion serves the marketing

It would be possible to read the preceding glossary as an indictment of marketing departments and an exoneration of regulators. That reading would be wrong. The confusion serves multiple actors, and the regulatory apparatus has its own contributions to it.

The first contribution is structural. The FDA’s regulatory vocabulary, developed across forty years of statutory amendments and agency rulemaking, is genuinely complex. Clearance and approval mean different things because they describe different processes that were created at different times under different statutes. The agency did not invent the words to confuse anyone. The words evolved to describe specific procedures. A consumer who has never read a regulatory document is not equipped to disentangle the vocabulary, and the agency does not, as a matter of mission, run consumer education campaigns sufficient to fix the gap.

The second contribution is competitive. Companies that have invested in the more rigorous pathways, De Novo with prospective clinical evidence, or PMA with full clinical trials, have a strong commercial interest in the public understanding the distinction. Companies that have not, and that hold only 510(k) clearance based on retrospective analytical validation, have an equally strong interest in the public not understanding it. The result is a marketplace in which the strongest evidence based products and the weakest evidence based products both describe themselves as cleared by the FDA, and consumers are left to navigate the difference on their own.

The third contribution is the amplification chain we wrote about in this publication’s anchor essay, The Literature Is a Debate, Not a Record. Press releases use FDA cleared and FDA authorized interchangeably. News articles, working from press releases under deadline pressure, repeat the imprecision. By the time the language reaches the consumer, the distinctions have been flattened. The amplification chain is not malicious. It is the predictable output of a system in which each link is optimizing for clarity at the cost of precision.

The fourth contribution is the rate of healthcare AI authorization itself. The FDA’s authorized AI device list has grown from a handful in 2015 to over a thousand by early 2025, with annual authorization rates that continue to climb. The volume of products available for confused marketing has expanded much faster than the consumer literacy required to navigate them. A reasonable forecast is that the gap will continue to widen for several more years.

A reader’s method

For the reader who wants to parse FDA language with any reliability, a short method is sufficient. When you encounter FDA approved in healthcare AI marketing, look for a PMA number or, if the product is a drug or biologic, a New Drug Application or Biologics License Application number. If neither is present, treat the phrase as inaccurate.

When you encounter FDA cleared, expect 510(k). Search the FDA’s 510(k) database by company or product name. Read the substantial equivalence summary. Note what the predicate device is, when it was originally cleared, and what evidence the new applicant submitted, whether that was retrospective analytical validation, a small bridging study, or something more substantial. The summary documents are short, generally a few pages, and are written in language a careful lay reader can navigate.

When you encounter FDA authorized, find out which authorization. De Novo, EUA, and umbrella usage are not equivalent. The De Novo Authorization Letter and the corresponding device classification order are public documents and worth reading. The EUA Letter, if applicable, will state explicitly the basis on which the authorization was granted and the conditions under which it remains in effect.

When you encounter FDA granted or FDA designated, default to assuming Breakthrough Designation. Verify on the FDA’s public Breakthrough Devices list. A designation is meaningful but distinct from clearance or approval, and a product that has only the designation has not been authorized for commercial sale on the basis of that designation alone.

When you encounter FDA registered, treat it as a negative signal. The phrase typically appears precisely when stronger phrases do not apply. A product that uses FDA registered in marketing has chosen the strongest accurate FDA related phrase available, which is usually the weakest meaningful one.

When you encounter the supplement disclaimer, the phrase not evaluated by the Food and Drug Administration, you have arrived at the regulatory honest accounting. The rest of the marketing copy on the page should be read in that light. Whatever the product claims, the agency has not reviewed those claims, and the claims should be evaluated on the strength of the underlying science alone, not on any regulatory endorsement.

Back to the contrast

The two products at the top of this piece, both legally describable as FDA authorized, illustrate the full range of what the phrase can hide. The diabetic retinopathy diagnostic was authorized on the basis of a prospective preregistered clinical trial. Its developer worked with the FDA for years to design the trial, set the predetermined sensitivity and specificity thresholds, and verify that the algorithm met them in a real primary care setting. The evidence behind the authorization is publicly available and methodologically strong.

The typical healthcare AI product cleared via 510(k) was authorized on the basis of substantial equivalence to a predicate, often with retrospective analytical validation on a test dataset that may or may not resemble the populations and clinical workflows in which the product will eventually be used. The evidence behind the clearance is, in many cases, not publicly available in a form that would let an independent reader evaluate it. The clearance is real. The clinical evidence underneath the clearance is, in most cases, weaker than the consumer’s intuitive reading of the phrase FDA cleared implies.

Both products operate legally. Both products may, in fact, work well in deployment. The question is not whether the products are legitimate, but whether the consumer reading the marketing has any reasonable way to distinguish between them. Under current marketing conventions, with current regulatory vocabulary, with the current amplification chain converting press releases into news articles into social media posts, the answer is mostly no.

This piece exists to close, by a small amount, the gap between the regulatory accounting and the marketing accounting. The reader who learns the six terms, internalizes the differences between them, and runs the short verification method on any healthcare AI product they encounter will see a different field than the field most consumers see. The strongest evidence based products will become recognizable. The weakest will become equally recognizable. The middle, which is most of the market, will become legible for what it is.

That legibility is the substantive deliverable of healthcare discovery. The terms are knowable. The pathways are documented. The evidence behind any given clearance can, in most cases, be looked up. The reader who does the work is no longer at the mercy of phrases they cannot parse.

Sources and further reading

US Food and Drug Administration. The 510(k) program: evaluating substantial equivalence in premarket notifications. Guidance document.

US Food and Drug Administration. De Novo classification process and decisions. Public documentation.

US Food and Drug Administration. Premarket approval (PMA) regulatory pathway. Public documentation.

US Food and Drug Administration. Breakthrough Devices Program. Public documentation.

US Food and Drug Administration. FDA in Brief: FDA calls on certain firms to stop producing and issuing misleading FDA Registration Certificates. March 3, 2021.

Abramoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI based diagnostic system for detection of diabetic retinopathy in primary care offices. npj Digital Medicine. 2018;1:39.

Digital Diagnostics. LumineticsCore (formerly IDx DR) FDA De Novo authorization documentation, April 2018.

Dietary Supplement Health and Education Act of 1994 (DSHEA). Public Law 103 to 417.

Analyses of FDA authorized AI and ML enabled medical devices reporting 94 to 97 percent of authorizations have come through 510(k), including landscape papers in npj Digital Medicine, Electronics, JACC: Advances, and a 2025 analysis of the 168 ML enabled Class II devices authorized in 2024.

Free Daily Briefing

The Latest Longevity Science.
Delivered Every Morning.

Join researchers, physicians, and health professionals getting daily breakthroughs in AI-driven medicine, epigenetics, and longevity research.

Support the research that powers this editorial

Subscribe Free

No spam. Unsubscribe anytime. We respect your inbox.