ARIA, the UK government agency that funds high-risk “moonshot” R&D, has selected 12 projects for funding in a competition focused on building “AI scientists,” systems intended to run an end-to-end scientific workflow. ARIA defines these systems as able to generate hypotheses, design and execute experiments (often via automated labs and robotics), analyze results, and potentially repeat the loop with minimal human intervention, with human researchers acting as overseers who set initial questions. The agency received 245 proposals and doubled the funding it originally planned to allocate due to the volume and quality of submissions. The grants are around £500,000, smaller than ARIA’s typical £5 million multi-year projects, and are designed as a fast, broad sampling of the state of the art to establish a baseline for future large-scale funding. Examples mentioned include Lila, whose team says the grant will support building and documenting a reproducible AI-robotics experimental loop around a focused scientific problem, and a London startup in stealth developing an AI scientist called ThetaWorld that uses large language models to design experiments related to battery-relevant physical and chemical interactions, with experiments run in an automated lab by Sandia National Laboratories in the US. ARIA’s leadership also notes the challenge of hype and the difficulty of evaluating frontier claims when results are shared via press releases rather than peer review.