US-EIN-91-1157127-400868 PATH AI Co-Scientists – agentic AI tools for scientific hypothesis generation coupled with domain-specific models (e.g., protein design tools) – are proliferating rapidly, but the field lacks robust ways of determining what works, when, and why, which risks wasting time, effort, and resource on tools that are not fit-for-purpose, ultimately slowing scientific progress. This programme aims to address this problem by creating: (1) a living landscape map of Co-Scientist tools, (2) an openly published benchmark (i.e., a suite of automated evaluations) to test the performance of Co-Scientist tools on a range of relevant tasks, and (3) a light-touch secretariat to help FCDO and its partners implement Co-Scientist workflows to accelerate discovery. In parallel, we will collaborate with FCDO’s extensive subject matter expert bench to demonstrate the potential impact of best-in-class AI Co-Scientists by applying them to address critical global health challenges: novel susceptibility testing for antimicrobial resistance (e.g., a low-cost and stable point-of-care susceptibility test suitable for use in LMICs); malaria vector control (e.g., a novel low-cost, synthetic mosquitocidal agent); and tentatively a third challenge to be agreed with FCDO. For each challenge, we will move from AI-assisted hypothesis generation to a small, catalytic validation study, such that by the end of the initial (2-year) phase, we have tangible evidence of success from at least one exemplar field, ready for further investment from other donors to drive global impact. PATH 24389.64 24389.64