For a decade, "cancer vaccine" was mostly a phrase in press releases. In 2026 it started to look like data. At the ASCO annual meeting in May, Moderna and Merck reported five-year results for a personalized mRNA vaccine in high-risk melanoma: added to the immunotherapy KEYTRUDA, it cut the risk of recurrence or death by 49% compared with KEYTRUDA alone [source: Merck, 2026]. Weeks earlier, BioNTech reported that in a small pancreatic-cancer study, seven of eight patients whose immune systems responded to a personalized vaccine were still alive four to six years after treatment [source: BioNTech, 2026]. Those are striking numbers for two of the hardest cancers to treat.
They are also easy to misread. So this article asks a narrower, checkable question: what has actually been shown, and what has not? The short version is that the early results are genuinely encouraging, that they come mostly from mid-size or small trials rather than the large confirmatory ones, and that — despite the headlines — no personalized cancer vaccine has been approved anywhere. Everything below is a therapy still under investigation.
Two things these vaccines are not
Before the results, two clarifications, because both confusions are common.
First, these are not preventive vaccines. The shots most people picture — HPV or hepatitis B — stop a virus that can later cause cancer, and you get them while healthy. The vaccines in this article are therapeutic: they are given to people who already have cancer, usually after surgery has removed the visible tumor, to train the immune system against any cancer cells left behind. In the trials, they are almost always paired with an immune checkpoint inhibitor, not given alone [source: Merck, 2026].
Second, this is not gene editing. A personalized cancer vaccine does not rewrite the patient's DNA. It works more like a wanted poster. As a tumor grows it accumulates mutations, and some of those mutations produce abnormal proteins — called neoantigens — that healthy cells never make [source: PMC review, 2024]. The vaccine shows the immune system pieces of those neoantigens so that T cells learn to hunt down cells carrying them. The mRNA is the delivery format, the same technology behind the COVID-19 shots, repurposed to carry a tumor's mutation signature instead of a virus's.
Why every vaccine is different
The word "personalized" is doing real work here. Because each tumor's mutations are unique, each vaccine is manufactured for one patient and no one else.
The workflow is roughly this. Surgeons remove the tumor; the tissue and a blood sample are sequenced to catalog the tumor's mutations; software predicts which of those mutations are most likely to be visible to that person's immune system; the top candidates are encoded into a bespoke mRNA vaccine and manufactured [source: PMC review, 2024]. Moderna's melanoma vaccine encodes up to 34 neoantigens per patient [source: Merck, 2026]. Start to finish, current workflows take roughly four to seven weeks from surgery to the first dose [source: PMC review, 2024].
That individualized production is also the field's central practical problem. A pill can be mass-produced; a per-patient vaccine cannot. Turnaround time, batch-to-batch variability, cost, and the regulatory question of how you approve a product that is different for every recipient are the obstacles most often named between promising trial data and routine use [source: ScienceDirect review, 2026]. The science of whether it works and the logistics of whether it can scale are separate questions, and both are unsettled.
The melanoma results: what the numbers say
The most mature evidence is in melanoma, from a trial called KEYNOTE-942. It enrolled 157 patients whose high-risk (stage III/IV) melanoma had been surgically removed, and randomly assigned them to Moderna's vaccine plus KEYTRUDA (pembrolizumab) or to KEYTRUDA alone [source: Merck, 2026]. When the primary results first read out, the combination reduced the risk of recurrence or death by 44% versus KEYTRUDA alone (hazard ratio 0.56), meeting the trial's main goal [source: The Lancet, 2024].
At five years, that benefit held up. With a median follow-up of about 60 months, the combination reduced the risk of recurrence or death by 49% (hazard ratio 0.51) and the risk of distant metastasis or death by 59% (hazard ratio 0.411) [source: Merck, 2026]. Those are durable, meaningful effect sizes.
But read the fine print, because it matters. Overall survival — whether patients actually live longer — was not established. That analysis was exploratory and immature, based on only 14 deaths, and the trend, while positive, was not statistically significant [source: Merck, 2026]. Recurrence-free survival is an important endpoint, but it is not the same as proof that the vaccine helps people live longer; that question is still open.
Two more layers of caution. KEYNOTE-942 is a phase 2b trial — mid-size, and designed to justify a bigger one, not to be the final word. The confirmatory phase 3 trial, INTerpath-001, enrolls roughly 1,089 patients and has finished recruiting, but its results are not yet in [source: Merck, 2023]. And these figures come from the companies developing the product — a legitimate reason for scrutiny — though the data have been published in peer-reviewed journals including The Lancet and presented at ASCO, which is a real point in their favor [source: The Lancet, 2024]. On the strength of the early data, the vaccine received the FDA's Breakthrough Therapy Designation in 2023 and the EMA's PRIME designation the same year [source: Merck, 2023]. Those are accelerated-review labels. Neither is an approval.
The pancreatic story: smaller, earlier, promising
The second headline comes from pancreatic cancer, which is far deadlier and where any signal is notable. Here the vaccine is BioNTech's autogene cevumeran, tested in a phase 1 trial of just 16 patients run at Memorial Sloan Kettering, combined with the checkpoint drug atezolizumab and standard chemotherapy [source: Nature, 2023].
The immunological result was clear: 8 of the 16 patients developed strong, vaccine-induced T-cell responses against their tumor's neoantigens [source: Nature, 2023]. On extended follow-up, those 8 responders had substantially longer recurrence-free survival than the 8 non-responders, whose cancer came back at a median of about 13 months (a difference reported as statistically significant, P = 0.007) [source: Nature, 2025]. At the latest update in 2026, seven of the eight responders were still alive four to six years out — remarkable in a disease where most patients do not survive that long [source: BioNTech, 2026].
The caveats here are even larger than in melanoma, and honesty requires stating them plainly. This is 16 people in a single-arm phase 1 study — a signal to chase, not a settled fact. More subtly, the responder-versus-non-responder comparison is a correlation, not proof of cause. Patients whose immune systems mounted a response may simply have had biologically different tumors — more immunogenic, and perhaps destined for a better outcome regardless of the shot. Only a randomized trial can separate "the vaccine caused the benefit" from "the patients who did well were always going to." That randomized phase 2 trial, run with Genentech, is underway, with data expected in 2026 [source: BioNTech, 2026].
Still investigational — and why that word matters
Step back from the two headline programs and the picture is consistent. The field is broad but early: as of a recent survey, the large majority of personalized neoantigen vaccine trials were still in phase 1, and no such vaccine has reached full regulatory approval [source: ScienceDirect review, 2026]. Moderna and Merck are running phase 3 trials not only in melanoma but in resected non-small-cell lung cancer, with kidney and bladder cancer studies behind them — and the lung trial's results are not expected until 2035 [source: Merck, 2023].
The deeper challenge is one reviewers of the field keep returning to: there is a persistent gap between a vaccine provoking an immune response and that response translating into consistent, durable clinical benefit [source: Cancer Cell, 2026]. A vaccine can reliably wake up T cells and still fail to change how long people live, because a tumor's surroundings can suppress those T cells, because the predicted neoantigens may not be the right targets, and because the weeks of manufacturing give an aggressive cancer time to evolve [source: ScienceDirect review, 2026]. Immunogenicity is necessary but not sufficient. That is exactly why the immature overall-survival data in melanoma, and the correlational design in pancreatic cancer, are not pedantic footnotes — they are the crux.
What to watch
Three signals will show whether 2026's optimism becomes standard care. First, the phase 3 readouts: INTerpath-001 in melanoma is fully enrolled, and it — not the phase 2b — is the trial that can confirm or deflate the effect, ideally with mature overall-survival data [source: Merck, 2023]. Second, the pancreatic randomized trial: if BioNTech's phase 2 reproduces the phase 1 signal under proper randomization, a correlation becomes something much closer to cause [source: BioNTech, 2026]. Third, manufacturing and access: even a clearly effective vaccine has to be built per patient in weeks, at a cost and reliability that a health system can absorb, and that logistical story will decide whether the therapy reaches beyond top research centers [source: ScienceDirect review, 2026].
The balanced read is neither "cancer is cured" nor "it is hype." What the measured evidence supports is more specific: personalized mRNA vaccines have produced durable reductions in recurrence in mid-size and small trials of melanoma and pancreatic cancer, always alongside other drugs, with real effect sizes and real limits — unproven survival benefit, unfinished confirmatory trials, and an unsolved scaling problem. That is a genuinely hopeful place to be. It is not the same as a finished one. Read the evidence, not the headline.