AIがゼロから16種のウイルスを設計 ― バイオセーフティ法は時代遅れか？
Stanfordの「Evo」AIが、実験室でのテストで増殖に成功した16種の新規ウイルスを設計し、AIが機能する生物学を設計できることを証明した。この成果は、計算上の病原体設計がほとんど規制されていないという規制の空白を露呈させている。
scientists at stanford and the arc institute used a generative ai model named evo to design complete viral genomes, successfully synthesizing 16 distinct viruses that replicate in bacteria, some faster than the natural phi x-174. trained on 9 trillion nucleotides and specific phage data, the model proposed 700,000 genomes, of which 285 synthesized sequences yielded 16 functional viruses, confirming ai can engineer robust biological tools. while the team excluded human-infecting virus data to ensure safety, the research highlights a disconnect between ai capabilities and government oversight, as current us policies restrict physical enhancement of pathogens but not the computational creation of novel biological agents. regulators like the nih admit purely computational design is exempt unless it targets known 'entities of concern,' creating a gap that experts fear could be exploited for designing harmful influenza variants or other threats.