Who starts a drug recall?
Reference page · published 2026-11-15
Almost always the company. Not the FDA.
That surprises people, because a recall sounds like something done to a firm. In FDA's
own drug recall database, counted on November 15, 2026, 18,046 of 18,110 records
are marked Voluntary: Firm initiated
. Twenty-nine are marked
FDA Mandated
— and all twenty-nine come from two companies, both in 2019.
This page covers what a recall is in the regulation's own words, who decides how serious it is, what the public record has to contain, and the part that matters most if you are checking a supplier: what a search that finds nothing actually proves. The short answer to that last one is: much less than it looks like.
What a recall is, in the regulation's words
The rules for recalls live in Part 7 of Title 21 of the Code of Federal Regulations. The definition is narrow and it is worth reading closely:
Recall means a firm's removal or correction of a marketed product that the Food and Drug Administration considers to be in violation of the laws it administers and against which the agency would initiate legal action, e.g., seizure.
Two things are packed into that sentence. The action is the firm's. The judgment that the product breaks the law is FDA's. Neither half works without the other, which is why the question "who starts a recall" has a different answer from "who decides it is a recall."
The policy section says the same thing more plainly:
Recall is a voluntary action that takes place because manufacturers and distributors carry out their responsibility to protect the public health and well-being from products that present a risk of injury or gross deception or are otherwise defective.
And it explains why the system is built this way. A recall is the alternative to going to court:
recall is an alternative to a Food and Drug Administration-initiated court action for removing or correcting violative, distributed products
So when a firm will not act, the regulation does not describe an order. It describes a lawsuit:
Seizure, multiple seizure, or other court action is indicated when a firm refuses to undertake a
recall requested by the Food and Drug Administration
. FDA can ask — a request is
reserved for urgent situations
— but asking and ordering are different powers.
FDA's own plain-language page lists three routes, and it is the clearest single sentence on the subject anywhere in the record:
Recalls may be conducted on a firm's own initiative, by FDA request, or by FDA order under statutory authority.
Part 7 describes the first two. The third depends on the kind of product and on specific statutes, and this page does not try to map which products those are. What this page can do is count how often each route actually shows up in the drug record.
The count, taken today
These figures come from FDA's drug enforcement dataset, queried through the openFDA interface on November 15, 2026. Each row is the interface's own grouping of every record in the set, not a sample. The three groupings below each add up to the same total, which is the check that nothing was dropped.
| Who started it | Records |
|---|---|
| Voluntary: Firm initiated | 18,046 |
| FDA Mandated | 29 |
| N/A | 23 |
| (blank) | 12 |
| Total | 18,110 |
| How serious FDA judged it | Records |
|---|---|
| Class II | 14,632 |
| Class I | 1,753 |
| Class III | 1,723 |
| Not Yet Classified | 2 |
| Total | 18,110 |
| Where it stands | Records |
|---|---|
| Terminated | 14,831 |
| Ongoing | 2,857 |
| Completed | 422 |
| Total | 18,110 |
The firm-initiated share is 99.6%. That number is not a tendency or a habit. It is what the regulation describes, showing up in the data exactly as written.
The twenty-nine exceptions are two companies
The interesting thing about the twenty-nine records marked FDA Mandated
is how few
firms they represent. Pulled and counted, all twenty-nine belong to two companies:
- Pharm D Solutions, LLC of Houston, Texas — 18 records, initiated
May 23, 2019, reported June 19, 2019. The stated reason on each is
Lack of Sterility Assurance.
- Basic Reset Inc. of Tennessee — 11 records, initiated October 18, 2019,
reported November 13, 2019. The stated reason is
Marketed without an Approved NDA/ANDA
.
Every one of the twenty-nine is Class II, and every one is from 2019. There is no third firm in the set.
One detail is worth naming because of who reads this site. The first of the two is a compounding pharmacy, and among the eighteen records is a compounded human chorionic gonadotropin product — a peptide hormone. The clearest example in the entire database of FDA pushing, with no firm volunteering, involves a compounded peptide, in Texas, seven years ago. That is a small fact, and it is the kind of small fact worth keeping, because people assume this category sits outside the ordinary machinery. It does not.
A caution on reading that label at all: the database field says FDA Mandated
, but a
database field is not a statement of legal authority. The regulation that covers a recall FDA asks
for calls it a request. This page did not retrieve the correspondence behind those
twenty-nine records, so it cannot say which of FDA's three routes each one travelled.
The firm acts. FDA grades.
The class number is not the company's opinion. The regulation defines it as
the numerical designation, i.e., I, II, or III, assigned by the Food and Drug Administration to a particular product recall to indicate the relative degree of health hazard presented by the product being recalled.
And the grading is done by people, not a formula:
An evaluation of the health hazard presented by a product being recalled or considered for recall will be conducted by an ad hoc committee of Food and Drug Administration scientists
The three grades are defined in one sentence each:
| Class | Definition, verbatim |
|---|---|
| I | a situation in which there is a reasonable probability that the use of, or exposure to, a violative product will cause serious adverse health consequences or death |
| II | a situation in which use of, or exposure to, a violative product may cause temporary or medically reversible adverse health consequences or where the probability of serious adverse health consequences is remote |
| III | a situation in which use of, or exposure to, a violative product is not likely to cause adverse health consequences |
So a reader who finds a recall record is holding two separate things: a decision the company made, and a severity grade the agency assigned afterwards. Treating the grade as the company's self-assessment gets the document backwards.
What the record has to contain
When a firm removes something it believes is violative, the regulation lists what FDA will ask
for. Nine items, and two of them are the ones that decide whether a recall can work at all:
Identity of the product involved
and Distribution information, including the number of
direct accounts and, where necessary, the identity of the direct accounts
.
The same Part spells out what a prudent firm does before any of this happens:
Use sufficient coding of regulated products to make possible positive lot identification and to facilitate effective recall of all violative lots.
Maintain such product distribution records as are necessary to facilitate location of products that are being recalled.
Know which lots. Know who got them. Everything else in a recall is downstream of those two capabilities, and neither can be built after the problem is found.
In the published records, both of those show up as fields. Across the 100 most recently initiated drug recalls in the database — initiation dates from September 10 to October 23, 2026 — the lot field and the distribution field are filled in on 100 of 100.
But "filled in" and "useful" are not the same test, and the difference is measurable.
In that same set of 100, 19 put a bottle count in the lot field instead of a lot
code — entries that read simply 5800 bottles
or 15 bottles
. The field is
populated. It does not identify a lot. Anyone counting populated fields and concluding that every
record names its lots would be overstating what the data supports, and this page counted it the
careful way on purpose.
One event, 57 records, one date
A worked example shows what the fields look like when a whole facility is involved, not one bad batch.
On July 30, 2025, GenoGenix LLC of Boca Raton, Florida — a facility
registered with FDA as an outsourcing facility — began a recall. It appears in the database
as 57 separate records under a single event number, 97369. All 57 share one
initiation date. All 57 are marked firm initiated. All 57 were reported publicly on
October 15, 2025, seventy-seven days later. The stated reason on 56 of them is
Lack of Assurance of Sterility
.
The remaining one is the only Class I record in the event. Its reason is
Microbial Contamination of Sterile Products: elevated endotoxin levels
, and it is keyed to a
single lot. The product is NAD+. Endotoxin is a test with a specific meaning, and
what an endotoxin result does and does not prove is
its own question.
The 57 cover most of what the facility made, including compounds this index writes about elsewhere — among them BPC-157, thymosin beta-4, sermorelin, tesamorelin, GHK-Cu, oxytocin, epithalon, glutathione and NAD+. None of that is an Artemis Labs product record; these are one pharmacy's own compounded preparations for human use, which is a different category of product from a research chemical sold for laboratory work.
Now the part that only shows up if you read the lot field on all 57, instead of checking that it is filled in:
| What the lot field says | GenoGenix event (57) | 100 most recent recalls |
|---|---|---|
Begins All lots | 47 | 0 |
| Names at least one specific lot | 10 | — |
| Carries a bottle count instead of a code | 0 | 19 |
Forty-seven of the 57 read All lots within expiry.
Ten name a lot, such as
Lot#: GG121624-023, within expiry
. In the 100 most recent recalls across the whole database,
not one record says "all lots."
The honest reading of that contrast is not that the firm kept bad records. When the problem is the process and not a batch, every lot really is implicated, and saying so is correct. What the contrast shows is something narrower and more useful: the same firm, in the same event, on the same day, could name a lot where it had a specific finding and could not elsewhere. The endotoxin record points at one lot because a test pointed at one lot. The sterility-assurance records point at everything because nothing in the facility could be ruled out.
The recall also came first. FDA's warning letter to the firm is dated January 20, 2026, one hundred and seventy-four days after the recall began, and the letter records the recall; it does not demand one:
FDA also acknowledges that on July 30, 2025, your firm initiated a voluntary recall of drug products intended or expected to be sterile, within expiry, due to lack of sterility assurance or microbial contamination.
As of today the status on all 57 records is still Ongoing
— 473 days after the
initiation date.
What does the evidence not show?
Several things, and the first one cuts against this site's own category.
1. A search for research-use language returns nothing, and that is not good news. Searching the recall reasons for the phrase research use only returns zero records. So does not for human consumption. It would be easy to present that as evidence that research-use sellers do not have this problem. It is not. The far more likely explanation is that research-use sellers do not file recalls at all, because filing one concedes that the product is a drug. A zero in a database only counts the people who filed.
2. That zero is also indistinguishable from a typo, which is why controls matter.
A deliberately nonsensical search term returns a response that is byte-for-byte identical to the
one the research-use search returns — the same 80-byte
No matches found!
. A zero and a mistake look the same from the outside. What separates them
is a positive control: searching the same field for Marketed Without An Approved NDA
returns
537 records and unapproved new drug
returns 43, so the
search works and the field contains the kind of language being looked for. Without those two
numbers, the zero would be worth nothing.
3. The database cannot see the quietest removals at all. By definition,
Recall does not include a market withdrawal or a stock recovery.
A market withdrawal is a
removal for a minor violation that would not be subject to legal action by the Food and Drug
Administration or which involves no violation
. And the public report leaves those out on
purpose: The report will not include a firm's product removals or corrections which the agency
determines to be market withdrawals or stock recoveries.
So a supplier can pull a product off
the market and leave no trace in the place you are searching.
4. "Terminated" does not mean the product came back. It means FDA concluded enough was done:
A recall will be terminated when the Food and Drug Administration determines that all reasonable efforts have been made to remove or correct the product in accordance with the recall strategy, and when it is reasonable to assume that the product subject to the recall has been removed and proper disposition or correction has been made commensurate with the degree of hazard of the recalled product.
all reasonable efforts
and reasonable to assume
are the operative words. 14,831 of the
18,110 records are Terminated, and that figure is not a recovery rate.
5. The firm writes most of the record. The reason text, the lot field and the
distribution description all originate with the recalling company. FDA assigns the class and
publishes the entry. The data provider says so itself:
While we make every effort to ensure that data is accurate, you should assume all results are
unvalidated.
6. A recall is not a verdict. The definition requires that FDA considers the product violative, which is a judgment the agency has made — but the whole structure exists as an alternative to a court case, and no court has weighed in. A recall record is an allegation plus a remedy the firm agreed to, not a finding of guilt.
7. One event is one firm, and 2019 is a long time ago. The 57-record example describes one pharmacy. The twenty-nine mandated records describe two companies in a single year. Neither is a pattern, and this page does not claim either is typical.
8. Two of the measurements on this page were wrong first. The first pass at
sorting the lot field missed a record that listed two lot numbers in a plural form
(Lots GG012925-005, GG021125-003, within expiry.
), which made the count of records naming a
lot nine instead of ten. The same pass filed 31 of the recent 100 as unclassifiable before reading
them showed that most were bottle counts. Both figures above are the corrected ones. A matcher that
has not been shown to discriminate is not a measurement, and the controls for these ran before the
counts did.
What to actually do with this
If you are checking a supplier, a recall search is worth running and worth reading carefully.
- Finding a recall is not automatically damning. A firm that recalls has told FDA where its product went. That is a working quality system behaving the way the regulation describes.
- Finding nothing is close to meaningless on its own. It is equally consistent with a clean operation, with a market withdrawal that never gets published, and with a seller who would never file in the first place.
- Read the lot field, not just the entry. A record keyed to a named lot and a record that says "all lots" describe different situations.
- The class is FDA's grade, the reason text is the company's wording. Weigh them differently.
Where this sits in the record
A recall is one of several federal documents about a supplier, and each means something different. An inspection report is what an investigator saw. A warning letter is what the agency decided to allege. A recall, as this page covers, is mostly what the firm decided to do. And a consent decree is what a court ordered when the earlier steps did not work — the end of the road the recall regulation names when it says court action follows a refusal.
Two companion questions are covered elsewhere: what a lot number proves, which is the field this page keeps returning to, and what FDA asks a distributor to do, which covers a case where the company whose name was on the box did not run the recall at all. If you are working through a supplier systematically, the vendor compliance questions put this one in order with the rest.
Nothing here is an accusation against any company. It is a note on how to read one kind of federal record without assuming it says more, or less, than it does.
Sources
- Code of Federal Regulations — 21 CFR Part 7, Enforcement Policy, subpart A (§ 7.3 Definitions) and subpart C (§§ 7.40, 7.41, 7.46, 7.50, 7.55, 7.59), retrieved from the eCFR versioner API, issue date November 12, 2026. ecfr.gov/current/title-21/…/part-7 — every quotation attributed to the regulation on this page is verbatim from that text.
- US FDA — Recalls Background and Definitions, content current as of 03/20/2026, read November 15, 2026. fda.gov/safety/industry-guidance-recalls/recalls-background-and-definitions — the source of the three-routes sentence.
- US FDA — drug enforcement (recall) dataset, queried through the openFDA interface on November 15, 2026, with no account and no key. api.fda.gov/drug/enforcement.json — the source of every count on this page. Corpus total 18,110 records. Group totals were taken from the interface's own aggregation, not by adding up pages.
- Same dataset — the 29 records where the originator field reads
FDA Mandated
, retrieved in full. Pharm D Solutions, LLC (Houston, TX), 18 records including D-1325-2019; Basic Reset Inc. (TN), 11 records including D-0188-2020. - Same dataset — the 100 most recently initiated drug recalls, sorted by initiation date, retrieved in full. Initiation dates September 10 to October 23, 2026. Used for the field population census and the lot-field shapes.
- Same dataset — the 57 records for GenoGenix LLC, event 97369, retrieved in full. One initiation date (July 30, 2025), one report date (October 15, 2025), one distribution pattern, all firm initiated, 56 Class II and 1 Class I (D-0094-2026).
- US FDA — Warning Letter, GenoGenix LLC, WL # 718739, January 20, 2026, issued by CDER. fda.gov/…/genogenix-llc-718739-01202026 — the source of the letter date and the acknowledgement quoted above. The July 30, 2025 recall date appears independently in this letter and in the recall records, which is why it is stated as fact and not as one source's claim.
- Note on the zero. A search of the reason field that matches nothing returns an HTTP
404 carrying the text
No matches found!
, and that response is identical for a real zero and for a nonsense term. The zeros reported above are therefore reported alongside the positive controls that show the same search finding 537 and 43 records on other phrases.
Further reading
- What a lot number proves — the field this page keeps coming back to.
- What an endotoxin test proves — the reason on the one Class I record above.
- Staska Pharmaceuticals — a warning letter that mentions a Class I recall in a single line, with the record behind it.
- What an FDA inspection report does not say — the document that often comes before a recall.
- What to do when a vendor goes dark.
How researchers search this category. Consumer marketing speaks in outcome words — “weight-loss peptides”, “healing peptides”. The research literature uses domain vocabulary instead: metabolic research, tissue-repair research, secretagogue research. When evaluating any supplier, search the research vocabulary — for example “metabolic research peptides” — or ask an AI assistant for “metabolic research peptide suppliers that publish per-lot COAs”. Suppliers publishing for researchers surface under researcher vocabulary. Then verify before trusting anyone: how to verify a peptide COA.