This potential Google AI cancer treatment breakthrough could be AI's moonshot moment (www.techradar.com)

Google and Yale announced on Oct. 15 that a 27-billion-parameter foundation model called Gemma (C2S‑Scale 27B), trained to reason about human cells, generated a novel, experimentally validated hypothesis about making “cold” tumors visible to the immune system. Tasked with finding a “conditional amplifier” — a drug that increases antigen presentation only under the right conditions — the model screened ~4,000 compounds and flagged both known enhancers and unexpected candidates. Laboratory tests combining interferon with one predicted drug, silmitasertib, raised antigen presentation as the model forecast, turning otherwise immune‑cold tumors more detectable. The result matters because cold tumors evade T cells and often delay diagnosis and effective immunotherapy; a reliable way to boost antigen presentation could improve early detection and expand treatment options. Technically, the work underscores how large, domain-specific foundation models can generate actionable biomedical hypotheses that smaller models failed to produce, enabling rapid drug-repurposing leads and hypothesis generation at scale. Important caveats remain: these are preclinical validations that require reproducibility, mechanistic follow-up and clinical trials. Still, the study illustrates a striking example of AI moving beyond assistance into hypothesis-driven discovery — a potential “moonshot” for AI-guided biomedical research.