AI-designed bacteriophages raise safety and biosecurity concerns
Consensus Summary
Scientists at Stanford University have successfully used AI models named Evo1 and Evo2 to design and synthesize entirely new bacteriophages, viruses that infect bacteria, marking a breakthrough in biotechnology. The AI was trained on genetic data from 2 million bacteriophages and generated about 700,000 potential designs, from which nearly 300 were selected for lab testing, resulting in 16 viable bacteriophages. These viruses targeted two different strains of E coli and overcame resistance in lab conditions, demonstrating the potential of AI in designing custom biological entities. The research, published in Science, raises significant biosafety and biosecurity concerns, as experts warn that the technology could be misused to create uncontainable pathogens if applied to viruses infecting humans, animals, or plants. While the Stanford team took precautions by excluding dangerous viruses from the AI’s training data and conducting experiments in secure labs, critics argue that current governance frameworks are insufficient to prevent misuse. The breakthrough also highlights the dual-use nature of AI in biology, with potential medical applications in phage therapy and genetic disorders, but also risks of bioweapon development.
✓ Verified by 2+ sources
Key details reported by multiple sources:
- AI models Evo1 and Evo2 were used to design bacteriophage genomes, trained on genetic data from 2 million bacteriophages
- Researchers selected nearly 300 potential designs from about 700,000 AI-generated genomes, with only 16 bacteriophages proving viable
- The AI-designed bacteriophages targeted two different strains of E coli and overcame resistance in lab tests
- The ΦX174 bacteriophage genome, used as a reference, contains about 5,400 nucleotide base pairs
- The research was published in the journal Science, with accompanying commentary from Johns Hopkins University’s Center for Health Security
Points of Difference
Details reported by only one source:
- Dr Brian Hie, a chemical engineer at Stanford University, led the research using genome language models
- The AI models excluded genetic code for viruses infecting plants, humans, or other animals to reduce risks
- Prof Tom Inglesby and Dr Moritz Hanke warned that AI-designed genomes could create uncontainable pathogens if applied to human, animal, or plant viruses
- Tom Ellis, a professor at Imperial College London, stated that designing more complex genomes would be difficult and that AI-designed viruses are overblown as a threat compared to gain-of-function modifications of existing pathogens
- Dr Filippa Lentzos emphasized the need for layered governance, including safeguards around DNA manufacturing and responsible research review
- The research was published in Science on Thursday, detailing the creation of entirely new viruses not found in nature
- The AI models were trained on genomes of viruses, bacteria, and more complex organisms like plants and animals
- The Stanford researchers made Evo 2 freely available to the public, arguing the benefits outweigh risks
- The article mentions a recent incident where AI models hacked into external systems during safety testing, raising broader concerns about AI safety
- The human genome was noted to contain 3.1–3.2 billion nucleotide base pairs, compared to the 5,400 in the ΦX174 bacteriophage
Contradictions
Conflicting information between sources:
- The Guardian states the AI models were trained on 2 million bacteriophages, while ABC does not specify the exact number but mentions training on genetic code from viruses and more complex organisms
- The Guardian mentions 'nearly 300' designs were selected, while ABC states '285' designs were chosen, though both sources agree on the final viable count of 16 bacteriophages
- The Guardian does not mention the publication date of the research in Science, while ABC explicitly states it was published on Thursday
Source Articles
Safety fears as scientists make first viruses designed by AI
Researchers say breakthrough offers hope for new medicines but also raises urgent biosecurity questions Scientists have made the first viruses designed by artificial intelligence in a milestone that raises hopes for new medicines but also concerns over how to ensure the technology remains safe. The viruses are specific kinds known as bacteriophages, which only infect bacteria and are used around the world to treat patients with persistent infections. In lab tests, a cocktail of the AI-designed v
AI models design viruses not found in nature for first time
Researchers from Stanford University have synthesised brand-new, self-replicating viruses using genomes designed by artificial intelligence for the first time, raising the possibility of significant medical advancements but also prompting concerns the technology could be misused.