In a Stanford laboratory, artificial intelligence crossed a threshold that evolution alone had previously held: the design of complete, functional viral genomes from nothing but learned patterns in the code of life. Sixteen AI-written bacteriophages successfully infected bacteria in petri dishes, confirming that the gap between imagining a living thing and building one has narrowed in ways science is only beginning to reckon with. The achievement carries the dual weight of all transformative discoveries — a genuine promise to heal, and an equally genuine capacity to harm — arriving at a moment
AI designs first fully functional viruses from scratch in landmark breakthrough
The room spontaneously burst into applause
Mark
When you say the AI "designed" these viruses, what does that actually mean? Did it write code, or something else?
Mimi
It predicted genetic sequences the way a language model predicts words. The AI was trained on thousands of real genomes, learned the patterns, and then generated new sequences that followed those patterns but had never existed in nature. Then humans synthesized those sequences in the lab—turned the digital code into actual molecules.
Mark
And they worked immediately? Sixteen out of three hundred?
Mimi
Sixteen out of three hundred that were synthesized, yes. The researchers selected the most promising designs first, so the success rate was higher than random. But the point is they worked at all. That was the surprise. You can't predict whether a novel genome will actually function until you build it and watch it replicate.
Mark
Why is this different from what AI has already done with antibiotics?
Mimi
Antibiotics are molecules. Viruses are entire organisms—they have to replicate, interact with host cells, maintain their own integrity. It's orders of magnitude more complex. This is the first time AI designed a complete, functional genome from scratch.
Mark
The biosecurity people seem genuinely worried. Are they being alarmist?
Mimi
No. The same technology that designed these harmless phages could design something that infects humans. The researchers took precautions—excluded dangerous pathogens from training data, worked in secure labs. But those are choices, not laws of physics. Future researchers might make different choices, or work in places with less oversight.
Mark
So what's the actual risk? Could someone use this to create a pandemic?
Mimi
Theoretically, yes. But it would still require significant expertise and resources. The real concern is that as the technology becomes easier to use, the barrier to entry drops. Right now it requires a Stanford lab. In ten years, it might require a smaller team. In twenty, who knows.
Mark
What would actually stop someone from doing that?
Mimi
Regulation, oversight, international agreements—the same things that theoretically prevent nuclear proliferation. They work imperfectly. The researchers argue that existing safeguards go a long way, but they're also saying we need to think hard about this now, before the technology is everywhere.
Il Polso
- For the first time, AI models trained on genetic sequences generated complete viral genomes that actually worked — not simulations, but living, replicating viruses that consumed bacterial colonies in real laboratory dishes.
- The same computational logic that predicts the next word in a sentence was repurposed to predict the next genetic letter, producing 302 candidate virus designs, sixteen of which proved fully viable.
- Antibiotic-resistant infections kill tens of thousands annually, and the ability to computationally design bacteriophages — viruses that hunt bacteria without harming human cells — could offer a powerful new front in that fight.
- Biosecurity experts at Johns Hopkins immediately raised alarms: if AI can design viruses that work, the critical question is no longer whether the technology exists, but whether it can be kept from those who would use it to cause harm.
- The researchers built in precautions — excluding human pathogens from training data, working only with bacteria-targeting viruses — but those safeguards are voluntary, and the regulatory landscape across nations remains uneven and fragile.
In a Stanford laboratory, artificial intelligence crossed a threshold that evolution alone had previously held: the design of complete, functional viral genomes from nothing but learned patterns in the code of life. Sixteen AI-written bacteriophages successfully infected bacteria in petri dishes, confirming that the gap between imagining a living thing and building one has narrowed in ways science is only beginning to reckon with. The achievement carries the dual weight of all transformative discoveries — a genuine promise to heal, and an equally genuine capacity to harm — arriving at a moment when humanity's wisdom to govern such power remains unproven.
In a Stanford laboratory, researchers watched clear spots spread across bacterial colonies in petri dishes and understood they were witnessing something genuinely new: viruses designed entirely by artificial intelligence, alive and working. For the first time, scientists had used AI to write complete viral genomes from scratch, and sixteen of them functioned exactly as intended.
The method borrowed its logic from language models like ChatGPT. Two AI systems, Evo1 and Evo2, were trained on thousands of existing genomes — viral, bacterial, plant, and human — to predict the next genetic letter in a sequence, much as a language model predicts the next word. The team then asked the AI to generate bacteriophages: viruses that infect only bacteria and pose no danger to humans. From 302 candidate designs, sixteen proved viable. When clear zones appeared in dishes layered with E. coli — evidence the viruses were replicating and consuming their hosts — the room erupted in applause.
The medical implications are significant. Antibiotic-resistant infections kill tens of thousands of Americans each year, and bacteriophages offer a targeted alternative: viruses that hunt specific bacteria without harming human cells. The ability to design new phages computationally could open treatments for infections that have exhausted conventional medicine, and more broadly suggests a future where AI learns evolution's design principles and applies them to create drugs and therapies nature never produced.
But the same capability that promises healing also threatens harm. Biosecurity experts at Johns Hopkins published a commentary alongside the findings warning of urgent questions the technology raises — noting that while the Stanford team took careful precautions, including excluding human and animal pathogens from their training data, those safeguards are voluntary and depend on the choices of future researchers operating under varying levels of regulatory oversight around the world.
The scale of what was accomplished clarifies both the achievement and its limits. The AI-designed bacteriophages contain roughly 5,400 genetic base pairs; the human genome contains three billion. Designing simple living organisms from scratch, the lead researcher acknowledged, would require considerable further work — but is not impossible, and the team is already interested in attempting it. What happens next depends on how science and governments respond to a moment when the barrier between designing life and synthesizing it has, for the first time, genuinely collapsed.
In a Stanford laboratory, researchers watched petri dishes under early morning light and saw something no one had engineered before: clear spots spreading across bacterial colonies, evidence that viruses designed entirely by artificial intelligence were alive and working. The moment marked a threshold in synthetic biology. For the first time, scientists had used AI to write complete viral genomes from scratch, and sixteen of them functioned exactly as intended—infecting bacteria, replicating, doing what viruses do.
The breakthrough emerged from a deceptively simple idea borrowed from language models like ChatGPT. Just as those systems predict the next word in a sentence, the Stanford team trained two AI models called Evo1 and Evo2 to predict the next genetic letter in a sequence. The training data came from thousands of existing viral, bacterial, plant, and human genomes—a library of life's instruction manual. The researchers then asked the AI to generate something new: bacteriophages, viruses that infect only bacteria and pose no danger to humans. From 302 candidate designs the AI produced, the team synthesized them in the lab. Sixteen proved viable. Samuel King, a PhD student in the lab, described the moment of discovery with the kind of wonder that marks genuine scientific surprise. The phage were placed on dishes layered with E. coli bacteria, and the team waited. When clear zones began appearing—evidence the viruses were consuming their bacterial hosts—the room erupted in applause.
The implications ripple outward in two directions. In one direction lies genuine medical promise. Antibiotic-resistant infections kill tens of thousands of people annually in the United States alone. Bacteriophages offer a potential alternative: viruses that hunt specific bacteria without harming human cells. The ability to design new phages computationally could accelerate treatment for infections that have exhausted conventional medicine. More broadly, the technology suggests a future where AI learns the design principles evolution has encoded into life itself, then applies those principles to create drugs, enzymes, and therapies that nature never produced. Brian Hie, the Stanford assistant professor leading the work, frames it as a turning point toward "massively improving human health."
But the same capability that promises healing also threatens harm. The moment AI can design functional viruses, the question shifts from whether such technology will exist to whether it can be controlled. Thomas Inglesby and Moritz Hanke, biosecurity experts at Johns Hopkins University, published a commentary alongside the Stanford findings warning of "urgent biosafety and biosecurity questions." They noted that viruses designed to infect complex organisms—including humans—should never be pursued, yet the underlying technology makes such designs theoretically possible. The researchers themselves took precautions: they excluded human and animal pathogens from their training data, worked only with bacteriophages, and conducted all experiments in secure laboratory facilities. These safeguards matter, but they are not foolproof, and they depend on the choices of future researchers working in countries with varying regulatory oversight.
The scale of what was accomplished is worth understanding precisely. The bacteriophages designed in this study contain roughly 5,400 genetic base pairs—the letters of the genetic code. The smallest living cell contains around 500,000. The human genome contains three billion. Hie acknowledged that designing simple living organisms from scratch would require "a lot of work, but not impossible." The team is already interested in attempting it. Marc Güell, a synthetic biologist at Pompeu Fabra University in Spain, called the work "a very significant turning point" because humanity is now beginning to design biology on computers rather than discovering it in nature. Patrick Cai, chair of synthetic genomics at Manchester, described it as an "important milestone" suggesting that AI is learning evolution's design principles and opening doors to AI-assisted genome writing.
What happens next depends partly on how the scientific community and governments respond. The technology is advancing rapidly. The safeguards exist but remain voluntary and fragmented. The potential to treat disease is real and urgent. The potential for misuse is equally real. The Stanford team has demonstrated that the barrier between designing life and synthesizing it has collapsed. The question now is whether humanity can maintain the wisdom to use that power well.
Citazioni salienti
This is the first time generative AI has been used to design a complete genome, it's something that can replicate and have other functions inside cells—this was new territory for us.— Brian Hie, assistant professor at Stanford University
The findings raise urgent biosafety and biosecurity questions about whether the technology can be used without enabling serious harm.— Dr. Thomas Inglesby and Dr. Moritz Hanke, Center for Health Security at Johns Hopkins University
Vuoi la storia completa?
Leggi l'originale su BBC News ↗
Technology