Researchers published Nature results showing RFdiffusion — an AI protein design model — can generate de novo single‑domain antibodies with atomic precision and confirmed binding poses by cryo‑EM. The team reported designs against multiple therapeutically relevant targets, demonstrating the model’s ability to build accurate antibody loops that historically challenged computational methods. Co‑authors include members of the Institute for Protein Design and Xaira Therapeutics founders; the work pairs RFdiffusion with a hypermutation system to optimize affinity and specificity. Cryo‑electron microscopy validation provided structural confirmation that several designs adopt the intended binding conformations. Nobel laureate David Baker, who led the underlying protein‑design efforts, cautioned that while the results are a major technical advance, computational antibody design remains an early‑stage tool that will require broader validation and integration into discovery pipelines. For readers: RFdiffusion is a generative AI model that builds protein structures conditioned on target epitopes — a niche but rapidly evolving approach to biologic discovery.