Scientists train an artificial intelligence system on existing DNA data libraries and then use it to generate “recipes” for viral genomes, according to multiple reports. The researchers feed the AI large collections of genomic sequences and related information, then prompt it to propose new viral genome designs. The study finds that a subset of the generated genomes are viable—meaning they can produce functional viruses—rather than being merely theoretical designs. Across the accounts, sixteen of the AI-generated viral genomes are described as viable, resulting in new viruses. The reports also emphasize that the viruses correspond to genetic forms not found in nature, implying the model is able to generate genome sequences that do not match known naturally occurring viruses in the researchers’ reference datasets. The outlets describe the work as lab-based creation of viruses using AI-generated genome designs, rather than naturally isolated pathogens. While the sources agree on the overall experimental approach and the number of viable genomes reported, they provide limited additional methodological detail in the excerpts provided.