Semaglutide, sold as Ozempic for diabetes and Wegovy for weight loss, has become one of the most widely used medicines in the developed world. Its maker, Novo Nordisk, holds a direct commercial stake in demonstrating that the drug is both safe and effective. ² Regulators on both sides of the Atlantic — the U.S. Food and Drug Administration and the European Medicines Agency's safety committee, the PRAC — police that claim, adding and removing warnings as evidence accumulates. ²

The central difficulty in answering whether long-term use is safe is that "long-term" is a moving target. The two flagship trials that anchor the safety case ran for defined windows: STEP 5 followed obesity patients for 104 weeks, or two years, and SELECT enrolled 17,604 overweight patients with established cardiovascular disease for a mean of 39.8 months — barely over three years. ² That is the longest prospective safety data available. Many people now taking these drugs for weight loss expect to use them for far longer.

Complicating any single verdict, the evidence pools three different patient groups. The STEP trials studied obesity without diabetes; SELECT studied overweight patients with established heart disease; and an older trial, SUSTAIN-6, studied type 2 diabetes. ³ Their baseline risks differ enough that safety findings do not transfer cleanly from one to another.

Start with what the controlled trials measured directly. In STEP 5, over two years, serious adverse events were actually less common in the semaglutide group than in the placebo group — 7.9% versus 11.8%. ¹ The arithmetic holds: 12 of 152 patients on the drug versus 18 of 152 on placebo. The much larger SELECT trial pointed the same way, with serious events reported in 33.4% of semaglutide patients versus 36.4% on placebo, a finding the manufacturer's press release and the New England Journal of Medicine report identically. ² A later Nature Medicine analysis of SELECT found fewer serious events with the drug across every body-weight category. ¹ On the specific fears that dominate public discussion, SELECT's investigators reported no excess in serious events from gastrointestinal disease, kidney failure, pancreatitis, cancer, or psychiatric disorders, and STEP 5 recorded no pancreatitis in either group at all. ²

The headline SELECT figures of 33.4% and 7.9% look four times apart, but they are not comparable. SELECT ran longer, in an older and sicker population, and captured adverse events more broadly than STEP 5. ³ Longer follow-up and higher baseline risk, not a different drug, explain most of the gap.

The consistent problem across trials was not catastrophic events but tolerability. Gastrointestinal side effects — nausea, vomiting, diarrhoea — affected 82.2% of semaglutide users in STEP 5 versus 53.9% on placebo, an excess of roughly 28 percentage points. ¹ These symptoms carried a real cost: in SELECT, gastrointestinal problems drove 10% of drug patients to quit versus 2% on placebo, a five-fold difference, and discontinuation for any adverse event ran at double the placebo rate — 16% versus 8%. ² A smaller retrospective study of 175 diabetes patients reported no severe events such as vomiting or optic neuropathy in either group, though as a single observational source this carries less weight than the trials. ¹

Weight loss itself was substantial and durable within these windows: -15.2% versus -2.6% at two years in STEP 5, and still -10.2% versus -1.5% at four years in SELECT. ² SELECT also delivered a concrete benefit beyond the scale, cutting major cardiovascular events from 8.0% to 6.5%, a 20% relative reduction. ¹

Open: No controlled trial in the evidence extends beyond four years, so cumulative adverse events that might emerge only after year three or four of continuous use are undocumented.; Whether the favorable serious-adverse-event profile seen in cardiovascular and diabetes cohorts holds in the general weight-loss population, which SELECT and SUSTAIN-6 did not study.

The trial picture is reassuring on serious events; the picture from outside the trials is noisier and harder to read. A CDC study published in April 2025 estimated that semaglutide contributed to nearly 25,000 emergency-room visits between 2022 and 2023, with more than 82% of those occurring in 2023. ¹ That concentration — more than 82% of visits in 2023 — indicates a sharp year-over-year increase as use scaled up. ¹ The figure reaches the public through People magazine's account of the study rather than the underlying paper, and it carries no denominator: without the total number of users, an ER-visit rate cannot be calculated or compared against other weight-loss treatments. ¹

Specific mechanisms behind some of those visits are documented. In September 2024 the FDA warned that Ozempic can cause ileus, a bowel obstruction. ¹ Because semaglutide slows the stomach's emptying, it also poses a surgical risk: a peer-reviewed journal described two case reports of patients inhaling stomach contents during elective surgery despite having fasted, attributed to delayed gastric emptying. ¹ That finding prompted a concrete, manageable response — the American Society of Anesthesiology now advises stopping weekly GLP-1 injections one week before surgery. ¹

Two further real-world figures signal concern but rest on interested parties. Novo Nordisk's own materials report that, as of March 2025, the FDA's adverse-event database held 767 cases tied to compounded semaglutide — copycat versions, not the branded product — of which 574 (75%) were serious, 176 (23%) involved hospitalization, and 14 involved deaths. ¹ These counts describe compounded products and, like the ER estimate, lack any exposure denominator, so they cannot be turned into rates. ¹ Separately, the plaintiff law firm Motley Rice reports 3,848 pending actions in the consolidated GLP-1 product-liability litigation as of July 2026. ¹ Litigation volume records allegations, not established causation, and this too comes from a single interested source. As one narrative review candidly notes, real-world long-term data are "still emerging" and largely built on insurance claims or anecdote. ¹

Open: The total number of semaglutide users underlying the ER-visit, FAERS, and litigation counts is not in the evidence, so none of these figures can be expressed as a population rate or benchmarked against alternatives.; Whether the compounded-semaglutide adverse events reported by the manufacturer reflect the drug itself or the quality of unregulated compounded products.

Where the real-world data are murkiest, the clearest signal comes from what regulators have actually done. Their record on semaglutide runs in both directions, which is itself informative. On the most emotionally charged concern — suicidal thoughts — the agencies converged on reassurance. In January 2026 the FDA requested removal of the suicidal-ideation warning from GLP-1 labeling after a comprehensive review of 91 placebo-controlled trials covering 107,910 patients found no increased risk. ¹ The EMA's PRAC had reached the same conclusion nearly two years earlier, finding the evidence did not support a causal link to self-harm. ¹ Two independent regulators, reviewing separately, agreed.

On other fronts the agencies have added scrutiny rather than removed it. In May 2023 the EMA raised a thyroid-cancer safety signal for GLP-1 drugs. ¹ Both the FDA and EMA labels note that semaglutide causes thyroid tumours in rodents while stating that the effect on humans is unknown; Ozempic's U.S. label carries a boxed warning for thyroid C-cell tumours and bars use in patients with a personal or family history of medullary thyroid cancer. ² In January 2025 the PRAC opened a fresh review into whether semaglutide raises the risk of NAION, a rare eye condition that can cause sudden vision loss. ¹

Regulators have also policed how the drug is marketed. In September 2025 the FDA issued a warning letter to Novo Nordisk, finding that a video from an Oprah television special created a misleading impression by minimizing GLP-1 risks. ¹ That the regulator found the company's own safety messaging inadequate cuts against any assumption that industry communication can be taken at face value.

Open: The NAION review opened in January 2025 has no reported conclusion in the evidence, leaving that eye-damage question formally open.; Whether the thyroid signal raised in 2023 has been resolved, escalated, or closed by either agency.

The durability question turns the safety debate on its head: the risk may lie less in taking semaglutide than in stopping it. Three independent sources — a narrative review, patient-experience journalism, and the primary STEP 1 extension trial — agree that patients regain roughly two-thirds of lost weight within twelve months of discontinuation. ³ The trial data quantify it precisely: participants regained a mean 11.6 percentage points of weight one year after withdrawal, which against a starting loss of around 17% works out to about two-thirds, confirming the estimate. ¹

That matters because most real-world users do stop. A review in Obesity Reviews found that 68% of patients had no adherence to semaglutide or the related drug liraglutide after one year, while one-year persistence — simply continuing to fill the prescription — ran at 40%. ¹ These two figures look contradictory but measure different things: persistence tracks whether a patient keeps the medication at all, while adherence tracks whether they take it as prescribed. Both can be true if many who continue are dosing irregularly. ¹

The combination frames the drug as a chronic-disease treatment rather than a cure. Weight loss holds while the drug is taken and largely reverses when it is not — a pattern the primary trial data present as an empirical finding, not evidence of any inherent flaw in the medication. ¹ But it means the relevant safety question for most people is the safety of indefinite use, which is precisely the horizon the trials do not yet cover. ³

Open: No study in the evidence follows continuous users for the five-to-ten-year horizon that indefinite maintenance would require, leaving cumulative long-term risk under sustained use unquantified.; Whether structured behavioral or staged-treatment approaches can preserve weight loss after discontinuation is not addressed by the available evidence.

The evidence supports not one answer but several partial ones, each addressing a different slice of the question. Laid side by side, they explain why honest observers disagree.

The first explanation holds that within controlled multi-year settings, semaglutide's safety is genuinely favorable. The strongest support is direct: serious adverse events were equal to or lower than placebo in both STEP 5 and SELECT, SELECT found fewer serious events across all weight categories and no excess in feared organ toxicities, and the drug reduced cardiovascular events. ⁴ No graded claim contradicts this within its stated window. Its limit is the window itself — nothing here speaks to year five or beyond. What would discriminate is controlled data extending past four years with a breakdown of late-emerging events.

A second explanation accepts that serious events are rare but argues that gastrointestinal intolerance imposes a large real-world burden the trials undercount. It draws on the 82% GI adverse-event rate, the doubled and five-fold discontinuation figures, the aspiration case reports, the ileus warning, and the 25,000 ER visits. ⁴ Against it stands SELECT's finding of no excess serious GI events. ¹ This reading is plausible rather than established, because its real-world figures lack the denominators needed to prove a population-level rate. Population incidence data for severe GI complications, with exposure counts, would settle it.

A third explanation reframes the whole issue around durability: because two-thirds of lost weight returns within a year of stopping and most users stop within a year, the practical safety question is that of indefinite use, which remains unstudied. ⁴ This is well supported by the discontinuation and adherence evidence and contradicted by no claim; what it lacks is the long-horizon outcome data to say whether indefinite use is safe.

A fourth explanation focuses on rare organ-specific risks — thyroid, retina, gallbladder — still under surveillance. It rests on the EMA thyroid signal, the rodent-tumour labeling, the boxed warning, the SUSTAIN-6 retinopathy finding of 3.0% versus 1.8%, the gallbladder excess of 2.8% versus 2.3%, and the open NAION review. ⁴ SELECT's finding of no excess cancer partly counters it. ¹ It remains plausible: these signals are real regulatory concerns but not yet confirmed population risks, and the retinopathy data come from a diabetes population distinct from weight-loss users. ¹

Open: No evidence discriminates between whether elevated discontinuation reflects manageable early GI symptoms that subside or a persistent burden that makes sustained use untenable for most patients.

The evidence forces a bounded conclusion rather than a clean verdict. Within the two-to-four-year horizon that trials actually cover, the claim that serious harm is elevated is not supported: across the two largest trials, serious adverse events matched or fell below placebo, the feared organ toxicities showed no excess, and the drug reduced cardiovascular events. ⁴ On suicidal ideation specifically, two independent regulators reviewing large datasets both cleared the drug. ² These are the firmest findings in the record, and they are declarative.

Equally forced, though, is the conclusion that the question the public is really asking — is indefinite use safe? — is not answered here. The longest controlled follow-up is SELECT's 39.8 months, and a candid narrative review concedes real-world long-term data are still emerging. ² Tolerability is a documented, not speculative, problem: discontinuation ran double the placebo rate and GI-driven quitting five-fold. ² And most users stop within a year, regaining most lost weight. ⁴

Several signals sit in genuine stalemate. The real-world harm figures — 25,000 ER visits, 767 compounded-product adverse events, 3,848 lawsuits — are serious on their face but rest on single sources without denominators, so they can neither be dismissed nor converted into rates. ³ The NAION and thyroid reviews are open. ² On present evidence these are unresolved, not resolved in either direction.

Of the pooled cohorts, the STEP obesity-without-diabetes population is the closest analog to the off-label weight-loss user, while SELECT (established heart disease) and SUSTAIN-6 (type 2 diabetes) are not — so even the favorable findings transfer only partially to the population the question asks about.

One SPECULATIVE observation, labeled as such and offered only as reasoning: if the pattern of two-directional regulatory action continues — the suicidal-ideation warning removed after review, the NAION review opened — then the safety label a decade from now is likely to differ from today's in both additions and deletions. ² That is an inference about regulatory behavior, not a claim about the drug, and the evidence does not establish it. What the evidence does establish is narrower and firmer: for appropriately selected patients under medical supervision, over two to four years, the documented balance of benefit and harm favors the drug, with intolerance as the leading real cost — and the longer horizon remains an open scientific question.