Most cancer drugs do not fail because the underlying science is wrong. They fail because the wrong patients end up in the trial - people whose tumors were never going to respond to the mechanism being tested. Pathos AI is built around that unglamorous, expensive fact.
The company combines large-scale multimodal patient data - clinical records, molecular profiles, and medical imaging - with a proprietary AI foundation model. The goal is to predict, before a trial begins, which patients are most likely to benefit from a given therapy. That prediction feeds into how Pathos selects drug candidates, designs adaptive trials, and decides early whether to keep going or stop.
Rather than selling software to pharma, Pathos develops and in-licenses its own oncology assets and advances them through a clinical-stage pipeline. The company describes itself as building one of the largest multimodal foundation models in oncology, drawing on millions of patient records.
It runs on a deliberately lean model: small, AI-enabled teams designing and running studies that would traditionally require large armies of coordinators. Fewer people, more data, faster answers - and, if the thesis holds, fewer failed trials.
