Google’s C2S‑Scale 27B foundation model, developed with Yale and based on Gemma, generated a novel hypothesis identifying an interferon‑conditional amplifier that could boost antigen presentation and convert “cold” tumors to “hot.” Laboratory experiments validated the model’s in‑silico hitset, and the finding was reported by Google and detailed in a preprint and blog post. The team simulated effects of ~4,000 drugs across immune contexts, prioritized context‑dependent amplifiers, and confirmed several hits in vitro. The work illustrates an emergent capability of large generative models to propose mechanistic, experimentally testable hypotheses for complex biological contexts. While AI‑driven hypothesis generation is not a replacement for wet‑lab validation, the study shows scaled models can accelerate target identification and reposition existing drugs for immuno‑oncology strategies.