siRNA did not fail in cancer. It was given the wrong job Authors: Brett Schrand Published: 2026 Twenty years after RNA interference entered the clinic, the scoreboard is lopsided: every approved small interfering RNA (siRNA) drug targets the hepatocyte, nearly all through the same GalNAc-ASGPR conduit.1 Cancer got there early and left early, leaving behind programs, CALAA-01, Atu027, ALN-VSP, and DCR-MYC, that produced tolerability data and almost no durable benefit.2 The verdict calcified quickly: siRNA is a delivery problem, tumors are a hard one, and so siRNA is not an oncology modality. That verdict is wrong, or at least incomplete. Delivery was a real obstacle, but not the decisive one: the decisive error was the job we assigned the molecule, and the current generation risks repeating it with better chemistry. Those programs were asked to silence KRAS, MYC, PKN3, VEGF, or RRM2, each conceiving siRNA as a better small molecule: shut down a driver, kill the malignant cell, dose to maximum tolerated exposure. That put RNAi in competition with kinase inhibitors and antibodies on the axis where they are strongest, and carried a fatal handicap. Knockdown is incomplete and heterogeneous across a tumor, and cell-autonomous cytotoxicity scales with the fraction of cells you reach, so silencing MYC in 30% of tumor cells does not treat the tumor but merely selects against a third of it. Cytotoxic pharmacology demands you reach nearly every malignant cell, the worst possible assignment for a modality whose distribution into solid tissue is imperfect. One recent readout underscores the point. siG12D-LODER, an extended-release implant against KRAS injected into the tumor by endoscopic ultrasound, published phase 2 results in early 20263,4: it missed the primary endpoint overall but showed, in the G12D/G12V subset, a promising though non-significant survival trend (22.7 versus 13.4 months). It removes the systemic-delivery excuse, releasing payload inside the lesion over weeks, yet still yields only a subset-level signal. For a target as buffered as oncogenic KRAS, meaningful benefit appears to demand silencing the driver by well over 90%, sustained over time and across nearly the whole tumor, a bar unreachable even with the drug placed directly in the mass.5 The lesson is not that delivery failed, but that even excellent delivery cannot rescue a task this demanding. The obvious rebuttal is that the tools have matured, and they have. Stereopure and chimeric backbones have pushed durability from days to months and extended exposure beyond the liver,6 and antibody-oligonucleotide conjugates have built on a decade of ADC engineering, site-specific conjugation, DAR control, linker stability, for nucleic-acid payloads. These are real advances but are almost entirely advances in the conduit. They do not touch the mandate: a better antibody delivers silencing to more tumor cells, yet “more” is not “nearly all,” and never will be in a heterogeneous, poorly perfused tumor. A second generation that uses better delivery to pursue cell-autonomous killing will meet the same ceiling the first did. The field has framed its central problem too narrowly. Three questions must be asked together: how do we get the siRNA into the tumor cell, how do we get it out of the endosome and into the cytoplasm where it can act, and, most neglected, which targets are worth silencing at all? The first two are delivery problems; the third is biology, and biology decides whether the delivery problems are worth solving. RNAi’s defining advantage was never potency; it is that RNAi can be directed against any transcript, whether or not the encoded protein offers a small-molecule pocket or a surface epitope. Against a cell you simply want to kill, that breadth adds little; other modalities already kill well. When trying to target a specific and prolonged immune response, it is decisive, because many prime targets are the ones nothing else can reach, and this is where the future of siRNA in cancer lies, along two lines. The first is to make the tumor visible to the immune system. Knocking down the machinery that shapes antigen processing, transcript surveillance, or edits RNA and remodels the immunopeptidome as a downstream effect, surfacing peptides the tumor would otherwise never display. Silencing suppressors of the antigen-processing and presentation machinery can raise the density of surface antigens, while silencing pathways that degrade aberrant transcripts can surface cryptic and neoantigenic peptides the immune system was never tolerized against.7 Here the arithmetic that doomed cytotoxic RNAi inverts: a rendered-visible subpopulation can prime a T cell response that spreads beyond the cells the drug reached, so partial, transient knockdown, fatal for a cytotoxic mandate, becomes adequate, even ideal. The second is to target the immune compartment itself. The tumor cell need not be the target at all: siRNA delivered to T cells, macrophages, or dendritic cells can silence the nodes that keep them dysfunctional, TGF-β signaling that blunts CD8 effector function,8 exhaustion and checkpoint programs, and the polarization that keeps tumor-associated macrophages immunosuppressive. Reconditioning the microenvironment this way has no small-molecule or antibody equivalent, and reaching only a fraction of these cells can still reset the local balance. Both share one logic: the drug triggers or unblocks an amplifying immune response, then steps out of the way, asking the siRNA to do nothing it does badly. None of this is easy. Tumor penetration, the binding-site barrier, endosomal escape below traditional benchmarks,9 and immune-subset targeting all remain unsolved. But they will not be solved while we measure the wrong thing. Preclinical RNAi oncology reports tumor growth curves when they should report the fraction of intended cells, tumor or immune that achieve knockdown and the durability of that knockdown. For most programs nobody knows that number. Without it, we will misattribute the next failures as we did the last, concluding again that RNAi fails in cancer, when what failed was the question we asked it. The first generation of siRNA oncology failed trying to be a cytotoxic, and the second should not try to be one at all. Its future is immunologic: reshape what the tumor shows the immune system, or rewire the immune cells that must respond to it. During the preparation of this work, the author(s) used Claude for editing the manuscript. The author(s) reviewed and edited the output as needed and take full responsibility for the content of the published article. Declaration of interests The author declares no competing interests. References