Scientists design AI viruses unknown in nature spark medical hope and biosecurity alarm

In a landmark study published on Thursday (August 6) in the peer-reviewed journal Science, researchers from Stanford University and the Arc Institute in California demonstrated that generative AI models can write coherent, functional genetic blueprints for bacteriophages—viruses that infect only bacteria—and that these AI-designed viruses can replicate inside living cells and kill antibiotic-resistant Escherichia coli strains.

The work marks the first successful creation of fully functional, replication-competent viruses whose entire genomes were generated by AI, establishing a new frontier where machine learning meets life sciences and raising profound questions about the governance of biological design.

How the AI-designed viruses were created?

The research team, led by chemical engineer Dr Brian Hie of Stanford University, employed genome language models called Evo1 and Evo2—systems analogous to the large language models that power AI chatbots but trained instead on genetic sequences.

According to the Science paper, the models were initially trained on DNA data from millions of organisms across all domains of life, and then fine-tuned on a dataset of about 14,266 Microviridae genomes (the family that includes the well-studied ΦX174 bacteriophage) to specialise in generating bacteriophage genomes.

To minimise biosafety risks, the researchers deliberately excluded genetic information from viruses known to infect humans, animals or plants from the AI’s training data. The models then generated thousands of potential viral genome sequences. From this vast pool, the team selected 285 of the most promising designs for laboratory synthesis.

These synthetic genomes were inserted into E. coli bacteria, which acted as biological factories, reading the genetic instructions and producing the newly designed bacteriophages.

The process proved highly inefficient as only 16 of the 285 candidates yielded viable, replication-competent viruses. However, these 16 AI-designed phages were entirely novel, possessing sequence patterns distinct from anything found in nature. Some variants carried between 67 and 392 new mutations compared to the closest natural ΦX174 strain, with one (Evo-Φ2147) differing enough to potentially qualify as a new species under certain taxonomic thresholds.

AI-designed phages open new front in fight against superbugs

The breakthrough holds significant promise for addressing the global crisis of antibiotic-resistant bacteria. Bacteriophages are already used in some countries as an alternative therapy for persistent infections when conventional antibiotics fail. However, bacteria can develop resistance to natural phages over time, limiting their long-term effectiveness.

In laboratory experiments, a cocktail of the AI-designed phages successfully infected and killed E. coli strains that had developed resistance to the natural ΦX174 phage. Some of the synthetic variants even outperformed the original parent strain in terms of bacterial killing efficiency. The researchers wrote in Science that the ability to “rapidly design” genomes and tune them for specific bacterial targets while overcoming resistance could “transform phage therapy” and “expand biotechnological toolkits”.

Brian Hie told media, “This is a next step in the complexity that’s designable by generative AI… 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.”

Biosecurity warnings: “The governance to safely steer it does not exist”

Despite the medical potential, the study has ignited a fierce debate over biosecurity. In an accompanying commentary in Science, Dr Thomas Inglesby and Dr Moritz Hanke from the Johns Hopkins Center for Health Security warned that the findings raise “urgent biosafety and biosecurity questions”. They wrote, “Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions. The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.”

While the Stanford team’s work involved only bacteriophages that cannot infect humans, the researchers acknowledged that the study demonstrates a critical principle: generative AI can now produce complete, functional viral genomes that replicate inside living cells. Inglesby and Hanke cautioned that research aimed at designing pathogens capable of infecting humans, animals or plants should not be pursued, warning that such organisms “might encode new pathogens that cannot be contained by existing countermeasures”.

The Stanford authors themselves stressed that the work raised “important biosafety, biocontainment and biosecurity considerations” and urged others designing whole genomes to “consult both safety and security professionals throughout the project”.

Experts urge layered biosecurity as AI-designed pathogen risks grow

Experts have called for nuanced, layered governance rather than panic. Tom Ellis, professor of synthetic genome engineering at Imperial College London, described the research as impressive but noted that bacteriophage genomes are among the smallest and simplest to engineer. “This is literally the smallest and easiest genome to make,” he said, adding that the achievement also highlighted the difficulty of designing more complex organisms.

Ellis warned that an AI system trained on dangerous pathogens could theoretically be used to design more harmful viruses, but he argued that fears of fully AI-designed pathogens are “very overblown” compared to the more immediate threat of gain-of-function modifications to existing pathogens. He said controls over access to genetic data and restrictions on manufacturing suspicious genomes could help mitigate risks, noting that “governments are working hard to do this already”.

Dr Filippa Lentzos, a reader in science and international security at King’s College London, emphasised that safeguards should not focus solely on AI models. “It’s important to see the bigger governance picture and not focus regulation solely on the AI model,” she said. “A layered approach makes more sense: safeguards around model development and access, responsible research review, synthesis screening, and established laboratory biosafety and biosecurity.”

The Science study, titled “Generative design of bacteriophages with genome language models” (DOI: 10.1126/science.aec2657), has been widely covered by major media outlets. Meanwhile, the breakthrough positions AI as a dual-use technology with transformative potential for medicine and biotechnology, but also with the capacity to lower barriers for creating novel biological threats. As Dr Hie’s team demonstrated, AI can now traverse sequence spaces unexplored by natural evolution, generating functional viral genomes with no precedent in nature.