A Microsoft‑led red‑team exercise demonstrated that structure‑preserving, AI‑designed protein sequences can evade traditional sequence‑based nucleic acid screening tools, prompting recommendations to adopt structure‑aware methods. The team generated tens of thousands of synthetic protein homologs derived from known toxins and showed many variants retained three‑dimensional folds similar to concerning proteins despite large sequence drift, undermining homology‑only screening. The researchers argue screening providers must incorporate structure prediction and new threat models as generative protein design becomes widespread. The study did not synthesize proteins but used computational folding to assess risk. Authors recommended strengthening biosecurity pipelines—combining sequence, structural, and functional assessments—so DNA synthesis providers and regulators can better detect emergent threats enabled by generative AI. The work adds technical pressure to policy debates on biosecurity governance and the scalability of screening standards for commercial synthesis providers.