AI-designed viral genomes move beyond theory at Stanford

Researchers at Stanford University have developed large language models trained on genomic data that can generate entirely new viral sequences. The models, initially designed to create novel proteins and DNA sequences, have been adapted to produce viable virus genomes that share structural similarities with naturally occurring pathogens. While all generated viruses remain closely related to existing biological models, their novel features would be difficult for such organisms to evolve naturally. The team has raised concerns about the potential for future systems to design viruses targeting vertebrates and humans, prompting consideration of defensive measures in advance of such capabilities.