Writing the genetic code for a living virus — and watching it turn into a working E. coli killer — is no longer science fiction. Stanford researchers say their Evo 2 AI model generated DNA sequences for nearly 300 phages, and laboratory tests confirmed 16 with particularly strong activity against E. coli.
Evo 2 wrote whole phage genomes from a single snippet
Evo 2 generates new DNA sequences from a small starting snippet of a phage genome. In this case, the researchers asked the model to produce an entire ΦX174 genome in one left-to-right pass — a complete viral blueprint generated in a single sweep, not assembled piece by piece.
The study centres on bacteriophage ΦX174, pronounced "FYE-ex-1-7-4," a virus that naturally infects E. coli. Phages like this are nature's original antibacterial agents; now an AI is designing new versions of them.
Why an AI-designed phage that kills E. coli matters
E. coli is everywhere — in the human gut, in food, in water — and some strains cause serious, hard-to-treat infections. Antibiotic resistance is a growing global crisis, and phages are increasingly studied as an alternative or complement to drugs.
What makes this Stanford result notable is the direction: the AI started with almost nothing and produced complete, functional-looking genomes. The 16 phages that killed E. coli in lab tests suggest generative AI can design biology that actually works when built.
The human story behind the Evo 2 model
Evo 2 was created by Brian Hie, an assistant professor of chemical engineering at Stanford and a Dieter Schwarz Foundation Stanford Data Science Faculty Fellow. The experimental work described in the paper was led by Samuel King, a bioengineering graduate student.
Their different roles reflect a wider shift in science: one researcher designs and trains the AI, another tests its creations in the physical world. The laboratory is where the model's ideas either survive or fail.
Confirmed facts vs what remains unclear
Verified from the research brief: nearly 300 phages were synthesised from Evo 2-generated DNA; 16 showed strong E. coli-killing activity in lab tests; the model generated a full ΦX174 genome in one pass; King led the experiments under Hie's supervision.
Not yet known: how these candidates compare with natural phages, whether any could be developed toward therapy, and what the AI's design logic reveals about phage biology. Those details await the full paper and independent scrutiny.
Risks and the balanced view
Generative AI in biology carries responsibility concerns. A model capable of writing phage genomes could, in principle, be pointed at other organisms — raising dual-use questions that the scientific community will need to govern carefully.
There are also practical limits. Activity against E. coli in a lab dish is a long way from a proven treatment. Phage therapy itself remains a young field, with open questions about stability, delivery, dosing and regulation.
AI-designed biology is the wider trend
Evo 2 sits inside a wave of generative models now being applied to molecular biology. The shift is significant: AI is moving beyond predicting what proteins and genomes look like, and starting to produce novel sequences that researchers can physically build and test.
This closes an important loop — AI generates a hypothesis, the lab validates it, and the results refine the next generation of models. Stanford's phage experiment is one of the clearest examples of that cycle so far.
What this means for readers
For most people, the immediate takeaway is straightforward: AI can now create functional biological agents, not just analyse existing ones. That capability carries both promise and risk, and it deserves public attention.
For students and early-career researchers, the field of AI-guided synthetic biology is opening up — blending machine learning, microbiology and protein engineering into a genuinely new discipline.
Where this research could go next
The logical next step is deeper characterisation of the 16 promising phages — how efficiently they kill E. coli, how stable they are under different conditions, and whether they offer any advantage over natural phages.
Longer-term movement toward phage therapy remains speculative at this stage. But the researchers have demonstrated that the pipeline works: AI designs, the lab builds, and biology answers.
Our Take
This is a landmark proof-of-concept for generative biology. The numbers are small — 16 working phages out of 300 — but the principle is large: an AI model wrote functional genetic code from a brief starting prompt. The mature judgment of this work will come through peer review, replication and independent testing. For now, it stands as one of the most tangible demonstrations yet that AI can create biology, not merely predict it.
Frequently Asked Questions
What is Evo 2?
Evo 2 is a generative AI model created at Stanford by Brian Hie that generates new DNA sequences from a small starting snippet of a genome. In this study, it produced complete ΦX174 phage genomes in a single left-to-right pass.
What is a bacteriophage?
A bacteriophage is a virus that infects and destroys bacteria. The study focused on ΦX174 (pronounced "FYE-ex-1-7-4"), a well-known phage that naturally infects E. coli.
How did Stanford researchers make the phages?
Evo 2 generated DNA sequences for nearly 300 phages, which the team then synthesised in the laboratory. Lab tests narrowed the candidates to 16 with particularly strong E. coli-killing activity.
Can these AI-designed phages treat infections in humans?
Not yet. The 16 phages showed strong antibacterial activity in laboratory tests only. Developing a treatment would require extensive further research into safety, stability, delivery and effectiveness in living systems.
