From Code to Life: Out of hundreds of thousands of generated designs, scientists synthesized 302 viral genomes and introduced them into E. coli. 16 synthetic viruses came alive replicated successfully, and lysed target bacterial cells.
• Beyond Natural Evolution: Some synthetic phages outperformed natural viruses, incorporating evolutionary motifs not seen in standard strains.
• A Weapon Against Superbugs: A cocktail of these Al-designed phages overcame resistant bacterial strains where natural had phages had failed.
• The Biosecurity Dilemma: As highlighted by biosecurity experts at the Johns Hopkins Center for Health Security, the capability to compose functional viral genomes with generative Al is now an established reality, underscoring the critical need for global biogovernance.
______________________________
* Transcript:
On August 6th 2026, science published a paper: an AI had written the DNA of a virus, and that virus came alive. This had never happened before.
The model is called Evo, and it works exactly like Chat GPT, except it predicts the next letter of DNA instead of the next word in a sentence. It learned from millions of genomes from across the whole tree of life. It learned the language of life, not ours. Then the team tuned it on one small virus called Phi X One Seventy Four, a bacteriophage, which means a virus that only attacks bacteria.
This one only infects E coli, and it can't touch humans. Eleven genes, about five thousand letters, the smallest possible arena for a very big test. They asked the model to write its own versions of that virus's genome, and it returned hundreds of thousands of designs. Scientists built about three hundred of the best ones and dropped them into bacteria. Sixteen came alive. They replicated, burst the cells, and killed the bacteria.
Some outran the natural virus. One of them used a protein from a distant branch of evolution. So the model wasn't just copying nature; it had made something nature never made.
Now the hopeful half: bacteria evolve resistance to antibiotics, and when that happens, the drugs stop working. Medicine has wanted a second option for years. A mix of these designed viruses beat three resistant strains of E coli, where natural phages had failed. So we may have a way to fight back once the antibiotics give out. But a warning was published alongside it.
The Johns Hopkins Center for Health Security wrote: the ability to compose viral genomes with AI now exists. The governance to steer it does not. The team left human viruses out of the training data. That was a choice, and every choice can be unmade.
The real question isn't what they built, it's who owns the platform next.