AI-designed viruses: medical promise meets biosecurity risk

Artificial intelligence has begun designing complete viral genomes, with Stanford University and Arc Institute researchers creating 285 AI-generated bacteriophage designs, 16 of which produced functioning viruses when physically tested. This marks AI's entry into biological design, not just analysis. The AI models, Evo 1 and Evo 2, learned patterns in DNA from thousands of related phage genomes and generated previously unseen variants of the ΦX174 bacteriophage. One AI-designed phage combined genetic changes that conventional engineering struggled to achieve, demonstrating AI's ability to identify multi-gene combinations across entire genomes. This capability offers significant medical opportunities, particularly for phage therapy against antibiotic-resistant bacteria, while raising biosecurity concerns about accelerating harmful biological engineering. Experts advocate for graduated, auditable access to AI capabilities and screening based on biological function rather than sequence similarity alone.