AI models were used to generate roughly 700,000 candidate genomes for a bacteriophage based on the existing ΦX174 virus, which infects and destroys E. coli. Researchers synthesized 285 of the most promising designs, inserted them into E. coli, and found 16 produced viable, replicating bacteriophages, with some proving more effective than the natural virus. The find is framed as exciting for its biotech potential but concerning given the obvious dual-use risks of AI-designed synthetic viruses.
Questions this post answers
Did AI models actually succeed in designing a virus that works in real life?
Yes, two AI models generated around 700,000 candidate genomes for a bacteriophage based on ΦX174, a virus known for destroying E. coli. Researchers synthesized 285 of the most promising designs and inserted them into E. coli, and 16 produced viable, replicating bacteriophages, some more effective than the natural original. Follow how AI capabilities intersect with biosecurity risk on daily.dev as the research keeps advancing.