The box at the top of the Ozempic label is the first thing a reader sees and the hardest to interpret. It is titled RISK OF THYROID C-CELL TUMORS, it names medullary thyroid carcinoma, and it appears in near-identical wording on eight prescribing documents in this class.
Reading all of them on the same day makes something visible that reading one cannot. The warning is more uniform than the evidence underneath it. Two products in this class have rodent tumour findings and different warning status; one of them has no boxed warning at all.
All figures below were read on 14 September 2026 from FDA prescribing information published through openFDA.
The same box, four different rodent files
| Molecule | Rats | Mice | Boxed warning |
|---|---|---|---|
| liraglutide (Saxenda, Victoza) | adenomas and malignant carcinomas | adenomas and malignant carcinomas | yes |
| semaglutide (Ozempic, Wegovy, Rybelsus) | adenomas at all doses; carcinomas in males | adenomas and a numerical rise in carcinomas | yes |
| tirzepatide (Mounjaro, Zepbound) | adenomas and carcinomas combined at all doses | not tumorigenic | yes |
| dulaglutide (Trulicity) | dose-related increase in C-cell tumours | not stated in the label | yes |
| exenatide (Byetta) | benign adenomas in females at all doses | — | no |
Two rows in that table are doing the interesting work.
Tirzepatide's mouse study is stated in the Zepbound label in one line: in a six-month carcinogenicity study in rasH2 transgenic mice at 1, 3 and 10 mg/kg twice weekly, tirzepatide "was not tumorigenic." Its rat study found the opposite — a statistically significant increase in adenomas in males at 0.5 mg/kg and above and in females at 0.15 mg/kg and above, and in adenomas and carcinomas combined at every dose tested — at exposures the label puts at 0.1 to 1 times the human exposure at the maximum recommended 15 mg weekly. One species, at and below clinical exposure, is enough for the box.
Exenatide went the other way. Byetta's carcinogenicity section reports benign thyroid C-cell adenomas in female rats at every dose tested. And Byetta has no boxed warning.
Why the exenatide result does not produce a warning, in the warning's own words
The reason is written into the sentence every other label uses. The boxed warnings say the tumours were dose-dependent and treatment-duration-dependent, at clinically relevant exposures. Both halves are load-bearing, and the exenatide data fail both.
Dose dependence: Byetta's female-rat adenoma incidences run 14%, 11% and 23% across the low, medium and high dose groups, against two control groups at 8% and 5%. The middle dose sits below the low dose. That is not a dose-response curve.
Clinical relevance: those doses produced systemic exposures 5, 22 and 130 times the human exposure at the maximum recommended 20 mcg per day. A tumour that appears at 130 times human exposure is a different finding from one that appears at 0.1 times it, and the labels of the boxed-warning drugs report exposure multiples at or below 1 repeatedly. Semaglutide's rat study is the extreme case: the adenoma increase was significant at all dose levels, and the lowest of those doses produced an exposure the label describes as below quantification.
So the box is not a statement that a drug caused tumours in animals. It is a statement about the shape of the animal finding, and the shape is checkable in each label's own Nonclinical Toxicology section. That distinction is invisible from the box itself, and it is why a page that says "every GLP-1 causes thyroid tumours in rodents" and a page that says "only some of them carry the warning" are both accurate.
The only human cases named anywhere belong to the oldest drug
This is the part with the fewest people in it and the most weight.
Novo Nordisk writes, in the Wegovy label and again in the shared Rybelsus and Ozempic tablets label: cases of medullary thyroid carcinoma in patients treated with liraglutide, another GLP-1 receptor agonist, have been reported in the postmarketing period, and the data in these reports are insufficient to establish or exclude a causal relationship between MTC and GLP-1 receptor agonist use in humans.
Two things follow. The human signal in this class is a liraglutide signal — the Saxenda label carries the same sentence about its own molecule. And it is cited in semaglutide's labels, by the manufacturer of both, as a class consideration.
Eli Lilly's labels do not contain it. Read the Zepbound and Mounjaro boxed warnings and section 5.1 end to end and there is no human-case sentence at all: rodent data, an unknown, a contraindication, counselling, and the monitoring paragraph. The same warning, resting on a visibly different evidence base.
Neither company is doing anything improper. Novo has postmarketing reports on a molecule it has sold since 2010; Lilly's molecule reached the market in 2022. The point is what it does to a reader who checks one label and generalises: the human evidence for this warning is thin, it is concentrated on the drug fewest current patients take, and it is explicitly described by the people who reported it as insufficient to establish or exclude anything.
A contraindication with no usable screening test
The last paragraph of every boxed warning in this class is the one most likely to be skipped and the most practically useful.
Routine monitoring of serum calcitonin or thyroid ultrasound, the labels say, is of uncertain value for early detection of MTC in treated patients. Such monitoring may increase the risk of unnecessary procedures, because calcitonin has low test specificity and thyroid disease has a high background incidence.
Regulators rarely tell prescribers that a test is not worth running. Here they do, in the same box that contraindicates the drug outright in anyone with a personal or family history of medullary thyroid carcinoma or MEN 2. The logic is coherent once the two are read together: the contraindication does the work at the front end, by keeping the highest-risk people off the drug entirely, and no screening programme is offered at the back end because none would perform well enough to be worth its false positives. The labels do keep one thread: a significantly elevated calcitonin — patients with MTC usually exceed 50 ng/L — or a nodule found on examination or imaging should be evaluated further.
What the labels do not say
They do not say the drugs cause thyroid cancer in humans. They do not say they do not. The sentence "it is unknown whether ... as human relevance of rodent thyroid C-cell tumors has not been determined" appears, with the molecule swapped, in every one of them, and an unknown that has been sitting in a label since 2010 is itself a finding: the class has been in wide use for sixteen years and the documents have not moved.
What has moved is the population. The molecules with the postmarketing human reports are the ones fewest people now take, and the molecules most people now take have the shortest exposure histories. That is the same reason the class's pancreatitis reporting concentrates on the oldest drugs, and the same reason the side-effects hub treats report counts as a measure of attention rather than of risk.
Sources
FDA prescribing information published through openFDA, all read 14 September 2026: Zepbound (effective 2026-08-28), Mounjaro (2026-07-29), Ozempic injection (2026-07-30), Rybelsus and Ozempic tablets (2026-01-30), Wegovy (2024-04-23), Saxenda (2026-02-25), Victoza (2023-02-06), Trulicity (2023-08-05), Byetta (2025-09-02) — boxed warnings, Warnings and Precautions section 5.1, and Nonclinical Toxicology section 13.1 in each.