According to MIT Technology Review, a research team from California has successfully designed bacteria-killing functional viruses using artificial intelligence (AI). The team claims this achievement is the “first complete genome generative design” and marks an early step in AI-designed life forms.
This research, conducted by scientists from Stanford University and the nonprofit Arc Institute, explains in a preprint paper how an AI system designed new genetic codes for viruses. The team then synthesized 302 DNA chains based on these designs and exposed them to E. coli. The results showed that 16 of these AI-designed viruses successfully self-replicated and destroyed the bacteria host.
“Seeing these AI-generated viral spheres with my own eyes was truly awe-inspiring,” said Brian Hie, head of the Arc Institute’s relevant laboratory. “These viruses were created right here in our lab.”
“Evo”: Learning from Millions of Viral Genomes
The core of the project is an AI system called “Evo”, which operates similarly to a large language model. However, instead of learning from text, Evo is trained on biological data. Specifically, Evo learned from approximately 2 million bacteriophage genomes, rather than books or articles. The task given to Evo in this study was to design a variant of the phiX174 bacteriophage. PhiX174 is a simple phage with only 11 genes and a DNA sequence length of about 5,000 bases.
Jef Boeke, a biologist from New York University Langone Medical Center, stated that although viruses technically do not qualify as “living organisms,” this research is an impressive first step in AI-driven life design. He noted that the AI system’s performance was “unexpectedly good,” and its design was innovative, making adjustments to gene sequences and arrangements that human scientists had never considered.
However, not everyone shares this positive view. J. Craig Venter, a scientist who pioneered synthetic DNA research, criticized the method as merely a faster version of trial-and-error experiments. His lab had previously synthesized cells through similar processes but found them inefficient, relying on manual literature searches to filter designs.
The Technology’s Potential and Risks
This technology holds significant potential applications. For years, doctors have been exploring the use of bacteriophage therapy to treat multi-drug-resistant infections. Additionally, viruses play a key role in gene therapy, serving as vectors to deliver new genes into human cells. AI-designed viruses could significantly enhance the effectiveness of both treatments.
However, the risks of this technology cannot be ignored. Although the research team intentionally excluded data related to human pathogens when training Evo, Venter expressed deep concerns about the potential dangers. He warned that using this technology to design viruses like smallpox or anthrax could lead to catastrophic consequences. “Extreme caution must be exercised in any virus enhancement research, especially when such research is random and its outcomes are unpredictable,” he emphasized.
Moreover, scaling this technology to live cells would present enormous challenges. For example, E. coli’s DNA is approximately 1,000 times the size of the phiX174 bacteriophage. Boeke cautioned, “Designing the genome of a live cell would be astonishingly complex, far exceeding the number of subatomic particles in the universe.”
Despite these challenges, Jason Kelly, CEO of Ginkgo Bioworks, advocates for making AI-designed cell research a national priority. He envisions the creation of automated laboratories that continuously test AI-generated genome designs and provide feedback to optimize the system. “Cells are the fundamental building blocks of all life, and achieving this goal would be a national scientific milestone,” Kelly stated. “The U.S. should ensure it leads the way in this field.”
References:
- MIT Technology Review, July 2025
- Preprint paper on AI-designed bacteriophages
