The boxed warning on these drugs is the single most alarming thing a first-time reader encounters, and it is also one of the most widely misread. It does not say the drug causes cancer in people. It says the drug caused a specific tumor in rodents, and that nobody has determined whether that finding transfers. The distance between those two sentences is where all the useful information lives.
What the rodents actually showed
Calcitonin-producing C-cells sit in the thyroid, and medullary thyroid carcinoma is the cancer that arises from them. In a two-year study in rats, semaglutide produced a statistically significant increase in C-cell adenomas at every dose level, and carcinomas in males at the higher doses, at exposures comparable to human treatment. In a two-year study in mice, adenomas increased in males and females at every dose.
Tirzepatide's program reads differently. Its two-year rat study also produced C-cell adenomas and carcinomas, but its mouse study used a six-month transgenic model rather than a two-year lifetime design, and the drug was not tumorigenic in it. Anyone comparing the two molecules on this point is comparing two different experiments, not two different results.
Why the effect is species-specific
This is the part that almost never makes it into a summary. The rodent tumors are not an unexplained toxicology signal. They are receptor mediated, and the receptor is distributed differently in different animals.
The foundational work localized the GLP-1 receptor to rodent C-cells and showed that agonists triggered calcitonin release, upregulated calcitonin gene expression and produced C-cell hyperplasia in rats, and to a lesser degree in mice. In humans and cynomolgus monkeys, receptor expression in thyroid C-cells was low, and the agonists did not activate adenylate cyclase or generate calcitonin release in primates. Twenty months of liraglutide dosing at more than sixty times human exposure produced no C-cell hyperplasia in monkeys. Mean calcitonin in patients treated for two years stayed at the low end of the normal range.[1]
A follow-up study closed the mechanism. Thirteen weeks of exposure raised plasma calcitonin and C-cell hyperplasia in wild-type mice, and produced neither in mice lacking the GLP-1 receptor. The same work checked whether the drug activated the RET proto-oncogene, the mutation that drives most human medullary thyroid cancer, and found that it did not.[2]
So the honest summary is narrow. A receptor-mediated effect in a species that expresses the receptor on those cells, absent in a species that barely does, working through a pathway other than the one that causes the human disease.
What the warning requires, and what it does not
The labeled consequence is a contraindication: neither drug is for anyone with a personal or family history of medullary thyroid carcinoma, or with multiple endocrine neoplasia syndrome type 2. That rule is absolute regardless of everything above, and it is the reason an intake has to ask.
Both labels then say something that surprises people. Routine monitoring of serum calcitonin, or routine thyroid ultrasound, is described as of uncertain value for early detection. The labels warn that such monitoring may increase the risk of unnecessary procedures, given the low specificity of the test and the high background rate of thyroid disease. A boxed warning that declines to recommend screening is telling a reader something about its own confidence.
The labels also note that medullary thyroid cancer has been reported after marketing in patients taking liraglutide, and that those reports are insufficient to establish or exclude a causal relationship.
The human epidemiology disagrees with itself
A French nested case-control study drew on the national insurance database, matching 2,562 thyroid cancer cases against 45,184 controls among people with type 2 diabetes. One to three years of GLP-1 receptor agonist use was associated with an adjusted hazard ratio of 1.58 (95% CI 1.27 to 1.95) for any thyroid cancer, and 1.78 (95% CI 1.04 to 3.05) for medullary thyroid cancer specifically.[3] That is a positive finding, and it should be reported as one.
The largest active-comparator cohort points the other way, and it reports absolute rates rather than only ratios. Across Denmark, Norway and Sweden, thyroid cancer occurred at 1.33 events per 10,000 person-years among people taking a GLP-1 receptor agonist, against 1.46 per 10,000 person-years on the comparator drug. The rate difference was −0.13 events per 10,000 person-years (95% CI −0.61 to 0.36).[4]
That is the figure worth holding onto, because it is the one in units a person can picture. Roughly one case per 7,500 years of treatment, and the difference between the two drug groups was smaller than the noise around it.
A 2026 target-trial emulation in a large United States records network followed adults starting these drugs for up to five years against usual care and a second drug class. It found modest increases in some non-malignant thyroid conditions among people with type 2 diabetes, and no increased risk of thyroid cancer in any group studied.[5]
How two studies reach opposite answers
Detection is the usual explanation and it is a good one. People who start a new diabetes or weight drug see clinicians more often, get imaged more often, and have more incidental thyroid nodules found. A case-control design comparing users against non-users absorbs that effect. An active-comparator design, which compares people starting one drug against people starting a different drug for the same condition, largely cancels it.
Duration is the other explanation, and it cuts against reassurance. Thyroid cancers are slow. The Scandinavian cohort's mean follow-up was under four years, which is short for an oncology question, and its own confidence interval leaves room for a meaningful relative increase on a very small base rate.
What this means before a prescription is written
For most people the practical content of this page is one question on a form. Has anyone in the family had medullary thyroid carcinoma or multiple endocrine neoplasia type 2. An intake that never asks it is skipping the only item on the label treated as absolute, which is part of what each provider review records against the criteria in the methodology.
One market-specific caveat. Most sellers here dispense compounded product, which does not arrive with the reviewed labeling quoted throughout this page, so the warning shaping it is the branded document rather than anything in the box. Price rather than paperwork is what usually decides where people buy — see the cost board — and commercial relationships are listed in the disclosures.
None of this is medical advice, and a family history question is one for a prescriber rather than a website.