What FDA approval actually means — and what it doesn't

"FDA approved" is doing a lot of work in a lot of sentences. It is a real bar, cleared by real evidence, and it is much narrower than most people picture.

The short version: approval says that for one specific use, in one defined population, over one defined period, a drug showed a treatment benefit that the agency judged to outweigh its known risks — on the evidence that existed on the day of the decision.

Every one of those qualifiers is load-bearing. This piece is about what each of them leaves open.

This is not medical advice. Nothing here is a reason to stop or change a prescribed medication — and stopping psychiatric medication abruptly can be dangerous. If you are in crisis, call or text 988 (US), 24/7, free.

What the approval bar actually is

The FDA publishes its own description of the trial stages, and the numbers are worth sitting with [1]:

Phase 3 is the pivotal stage — the one approval usually rests on. And the FDA states the limitation of everything before it plainly: "Phase 3 studies provide most of the safety data. In previous studies, it is possible that less common side effects might have gone undetected" [1].

So the safety picture at approval is built mostly from a few thousand people, studied for a period measured in months to a few years.

The four things approval does not say

1. It does not say the drug is better than the alternatives.

Approval generally rests on showing a treatment benefit against a comparator, and that comparator is very often placebo rather than the drug you are already taking. A newer medication can arrive on the market having never been tested head-to-head against an older one. "Approved" and "best available" are different claims, and only one of them is on the label.

2. It does not say the drug was tested for as long as you will take it.

This is the gap we counted ourselves. In the Resolv evidence library as it stood in August 2026: 235 verified records, 72 of them individual randomised trials. Fifty-seven of those 72 state how long they ran. Forty-seven of the fifty-seven measured their headline result at three months or sooner. Six ran past six months. Meanwhile the same library contains a meta-analysis pooling people who had been on benzodiazepines for about ten years.

The full count, by drug class, with denominators and the methodology, is here: how long were these drugs actually tested?.

That gap is not a scandal and it is not hidden. Running a ten-year placebo-controlled trial is close to impossible for practical and ethical reasons. But it does mean that "long-term use is supported by the approval evidence" is a sentence nobody can honestly say, and it is a fair thing to raise when a prescription is open-ended.

3. It does not say the drug works for you.

Trials report averages across a population that met the entry criteria. Those criteria routinely exclude people with the comorbidities, medication combinations, ages and circumstances that fill actual clinics. A statistically significant average effect in a selected sample is a real finding and a weak predictor of any individual result.

4. It does not say the safety picture is complete.

Here the FDA is more candid than most of its critics give it credit for. From the agency's own post-market page, verbatim [2]:

"Even though clinical trials provide important information on a drug's efficacy and safety, it is impossible to have complete information about the safety of a drug at the time of approval. Despite the rigorous steps in the process of drug development, limitations exist. Therefore, the true picture of a product's safety actually evolves over the months and even years that make up a product's lifetime in the marketplace."

Read that as the agency describing its own design. Approval is not the end of the evidence. It is the point at which the drug stops being studied in a controlled sample and starts being studied in everyone.

What happens after approval

The post-market stage is where a lot of what patients actually need to know gets discovered [2]:

• MedWatch — voluntary adverse-event reporting by clinicians, manufacturers and consumers.
• Manufacturer inspections — routine and for-cause, domestic and overseas.
• Post-market requirement and commitment studies — trials required of, or agreed to by, the sponsor after approval, gathering "additional information about safety and effectiveness, studying different populations and different dosages and using the drug in combination with other drugs" [4].
• The Sentinel Initiative — active surveillance across large electronic health-record and claims databases.

When this system finds something, the visible output is a Drug Safety Communication — the FDA telling clinicians and the public that what it knows has changed. New boxed warnings, restricted indications, and withdrawal requests all surface here.

The part that surprised us: where the old safety communications live

We went looking for the FDA's historical Drug Safety Communications while researching this piece. Here is what the agency's live page currently offers.

The page lists 41 communications, the oldest dated 14 January 2020 [3]. Everything published before that appears in exactly one place — a single bullet in the Resources section, which reads, in full:

"Full archived versions of Drug Safety Communications published before 2020"

The link on those words does not go anywhere on fda.gov. It goes to web.archive.org [3].

So the primary public record of what the FDA told the public about drug safety before 2020 — every communication from the era that produced a great many of the warnings now sitting on psychiatric drug labels — is served to you by a private non-profit library in San Francisco, because the agency links you there rather than hosting it.

We want to be careful about what that does and does not mean. It is not a cover-up. The FDA is doing the honest thing: it is telling you where the material is instead of quietly deleting the links. Federal web-archiving policy is a real constraint, not a plot. And the Internet Archive is a genuinely excellent library.

But it is a fact worth knowing about the durability of the record. Regulatory history — the trail of what was known, and when, and what the agency said about it — is not guaranteed to stay at a government address. If you are ever trying to reconstruct why a warning exists, expect part of the trail to run through an archive, and expect the agency itself to send you there.

For a mental-health-specific example of what lives in the recent, still-hosted portion of that list: the FDA required a boxed warning about serious mental-health side effects for the asthma and allergy drug montelukast (Singulair) on 4 March 2020 — an action taken decades after the drug's original approval [3]. That is post-market surveillance working. It is also a reminder of how long "working" can take.

What this does not mean

• It does not mean approval is worthless. The bar is real, most candidate drugs never clear it, and the FDA's own figures put the attrition at roughly 70% surviving Phase 1, about 33% moving from Phase 2 to Phase 3, and 25–30% moving on from Phase 3 [1].
• It does not mean unapproved products are safer. Supplements are not held to any of this. We wrote about what that actually means at supplements for mental health.
• It is not a reason to stop a medication. If this piece makes you want to change something, that is a conversation with your prescriber — with a taper plan and check-in dates agreed in advance — not a decision to make alone. Several psychiatric drug classes have documented withdrawal or relapse risks; we went through them at stopping antidepressants: withdrawal and relapse.

What to ask your prescriber

• What was this drug approved to treat, and is that what I have?
• Was it compared against placebo, or against another treatment?
• How long did the trials behind it run, compared with how long you expect me to take it?
• Have there been label changes or safety communications since it was approved?
• What would make us review this — a date, a symptom, a test?

The last one turns an open-ended prescription into a plan with a checkpoint, which is the single most useful thing a patient can do with any of this.

Sources

1. US Food and Drug Administration. Step 3: Clinical Research — The Drug Development Process. Phase tables and attrition percentages quoted as published. fda.gov — accessed 2026-08-25.
2. US Food and Drug Administration. Step 5: FDA Post-Market Drug Safety Monitoring. Source of the "impossible to have complete information about the safety of a drug at the time of approval" passage, quoted verbatim. fda.gov — accessed 2026-08-25.
3. US Food and Drug Administration. Drug Safety Communications. Page content current as of 06/10/2026; 41 communications listed, earliest dated 1-14-2020, latest 6-10-2026. The Resources section offers "Full archived versions of Drug Safety Communications published before 2020" as a link to `web.archive.org`. Montelukast boxed-warning communication dated 3-4-2020 appears in the live list. fda.gov — accessed and verified 2026-08-25.
4. US Food and Drug Administration. FDA's Drug Review Process: Ensuring Drugs Are Safe and Effective. Source of the postmarket requirement and commitment description. fda.gov — accessed 2026-08-25.

Related reading

• How long were these drugs actually tested? — the trial-duration audit
• Why your doctor gets paid for procedures, not outcomes
• The serotonin story: what the evidence said, and what patients were told
• Who funds psychiatric drug research?