Wednesday, 21 October 2026Support us

Aube.

News of progress
LabSingle source

AI designs working viruses that kill bacteria, in the lab

Languages for this article
Original · ENESFRITPT

Originally written in English. 5 languages available; yours is one click away.

Generated infographic: 16 — AI designs working viruses that kill bacteria, in the lab
AI-generated illustration

Coverage of the study reports that the researchers printed 302 AI-written genomes as DNA strands and mixed them with Escherichia coli bacteria. Some of them killed their targets. The work, led by Samuel King, a Stanford University PhD student, with the Arc Institute, has now been published in the journal Science.

The team did not start from nothing. It built on ΦX174, a natural bacteriophage, meaning a virus that infects bacteria, whose genome spans roughly 5,386 base pairs (the DNA letters) and encodes 11 genes. The AI models, called Evo, are genome language models, trained to read the structure of genetic sequences. The first Evo model, published in Science, was trained on 2.7 million prokaryotic and phage genomes. For the phage project, both models were then fine-tuned on around 15,000 genomes from the Microviridae family.

Generating a genome was only the first step. The team ran several rounds of computational and experimental filtering before printing anything. One of the designed phages, called Evo-Φ36, carried a shortened protein that did not function before when introduced into the natural virus. Coverage of the study reports 16 working genomes designed from scratch. A preprint had appeared on bioRxiv in September 2025, before the journal publication.

The biosecurity question follows directly from the technique. Johns Hopkins Center for Health Security researchers Thomas Inglesby and Moritz Hanke argue that the safety rules for this kind of work don't exist yet. Stanford and the Arc Institute have filed a provisional patent application covering the methods, and the inventors are named on it.

Biology labs gain a design tool: instead of editing an existing virus, a model proposes whole genome sequences to be built and tested. MIT Technology Review presents the work as a new way of seeing biology, and its reporting says AI-generated life "isn't yet" a reality, though it "could be next". No date for practical use, no price, and no application beyond bacteria has been announced so far.

16working genomes designed from scratch by AI, per coverage of the study

Sources — read the originals(Paris time)

MIT Technology ReviewEN

Pages consulted while writing

AI Designs Working Virus Genomes for the First Time — Artificial Science
Generative design of novel bacteriophages with genome language models
Generative design of bacteriophages with genome language models
Stanford, Arc Institute scientists design first AI-generated viruses
Stanford and Arc Institute: AI designs viable bacteriophages
AI-Designed Viral Genomes Raise Biosecurity Concerns
An AI Wrote 16 Working Virus Genomes From Scratch. They're Aimed at Bacteria, Not You, and That Distinction Matters a Lot.
AI Designed 16 Viruses No Nature Ever Produced
AI Designs Viable Bacteriophage Genomes, Combats Antibiotic Resistance
AI-designed viruses are here and already killing bacteria
0000

Read next

Comments

Loading the thread…

Sign in to leave a comment. Sign in