On 28 August 2026 BioNTech killed BNT122-01, a Phase 2 trial of autogene cevumeran (BNT122 / RO7198457), a personalized mRNA cancer shot built with Genentech. The patients had already had Stage II high-risk or Stage III colorectal tumours cut out. They still had circulating tumour DNA, leftover disease you can measure in blood. The trial asked whether the vaccine, given alone after surgery, beat watch-and-wait. An independent DSMB found a numerical imbalance in overall survival and said carrying on would not change the result. Genentech then confirmed the direction the company press release danced around: more deaths in the vaccine arm.

BioNTech still has not said how many died in each group.

What this product is, and is not:

This is not Comirnaty in a new box. Cevumeran is an individualised neoantigen vaccine: sequence the tumour, encode the mutations in mRNA, try to train T-cells on that patient's leftovers. Same firm, same class of molecule, different job. Conflating it with the COVID campaign is sloppy. Pretending the platform is therefore untouchable is sloppier.

Oncology trials fail constantly. Colorectal cancer after surgery, with an immunosuppressive tumour microenvironment, is a hard setting. Moderna and Merck just posted a Phase 3 win in melanoma, with a checkpoint inhibitor, not as monotherapy. BioNTech's own CMO called this result insight into "immunotherapy-insensitive" tumours. That is a real scientific sentence. It is also the sentence you issue when the survival curve went the wrong way.

Three facts sit badly with the culture of "trust the science" as a brand.

1. Futility was already crossed in October 2025.
The DSMB said the data were too immature to judge efficacy and there were then no safety objections, so BioNTech continued. European records, as reported, show recruitment going on for months after that line was crossed, including in Germany into April 2026. Continuing after futility is not automatically misconduct. It can be statistics. It is still a choice to keep putting people on an arm that had already failed its own "this is working" test. Ten months later the board was looking at deaths, not just recurrence.

2. "No new safety signals" next to "more deaths on the shot."
Those two phrases can both be true in regulatory English. Overall survival can worsen because the cancer progressed faster, because of chance in a modest Phase 2, because of something the product did that does not fit a pre-specified AE bucket. The public is not required to find that comforting. All-cause death is the endpoint that does not care about your codebook. If the company will not publish the counts, "no new signals" is a claim without a table.

3. The counts are still secret.
A survival imbalance serious enough to stop the trial is exactly the result that should be dumped as numbers, follow-up time, and causes. Investors got an 8% gap in the share price. Patients and doctors got adjectives. That is the same transparency habit critics spent five years describing in another mRNA program from the same house: announce the frame, withhold the cells.

It does not prove COVID mRNA shots kill net. Different construct, different population, different dose schedule, different risk. It does not prove personalised cancer mRNA is a dead field; combination trials continue, including pancreatic work with a checkpoint drug. Phase 2 imbalances can be noise. Until the Kaplan–Meier plots and death tallies exist, anyone who writes "the vaccine murdered the arm" is doing the mirror-image of the people who wrote "safe and effective" before the follow-up was in.

What it does prove is smaller and meaner. The platform is not a sacrament. An mRNA product from the company that taught the world that phrase can sit on the wrong side of overall survival, after a futility miss, while recruitment continues, and the first public sentence can still be process language instead of a body count.

"Safe and effective" collapsed two different questions into a bumper sticker: Does it do the job? and Does it pay for itself in harm? Regulators and companies then treated doubt about either question as an attack on the other. A cancer trial that fails efficacy and shows more deaths on treatment is what happens when those questions are forced apart again.

If cevumeran is merely the wrong tool for residual colorectal disease, publish the deaths and move the program to combinations. If the imbalance is small and messy, publish that too. What you cannot do, after 2020–2023, is expect the public to hear "numerical imbalance in overall survival," "more deaths in the vaccine arm," and "no new safety signals" as three harmonious notes.

The latest wrinkle is not that mRNA can fail. Every modality fails. The wrinkle is that the firm most identified with the slogan still will not put the deaths on the page, and that a trial already told it was futile was allowed to keep enrolling until survival, the only endpoint nobody can spin, pointed at the shot.

https://jonfleetwood.substack.com/p/biontech-kills-mrna-cancer-shot-phase