Not Just Coding and Medical Records! AI Can Now Design Entirely New Viruses to Help Eliminate 'Superbugs'

A U.S. research team has achieved a major breakthrough in biotechnology by successfully using generative artificial intelligence (AI) to design a completely new virus with full structural integrity and functionality—one capable of autonomous replication in the laboratory. This marks the first time in human history that an AI has designed a complete biological genome from scratch.

According to BBC reporting, the 16 newly created viruses produced in the lab are specifically engineered to infect and eliminate certain bacteria and pose no health threat to humans. Biological experts describe this achievement as a pivotal turning point in scientific history, potentially opening a new chapter in global disease treatment—though it also raises significant biosafety concerns.

In recent years, generative AI has rapidly advanced, helping engineers detect complex coding errors, assisting doctors in reviewing prescriptions and patient records, and even succeeding in designing novel antibiotics. However, constructing a fully functional, self-replicating virus from scratch presents a far more complex and difficult challenge than designing antibiotics.

Brian Hie, Assistant Professor at Stanford University, stated that this represents a new milestone in the high-complexity design capabilities of generative AI. The AI models used in this study, named Evo1 and Evo2, operate similarly to large language models like ChatGPT. The key difference is that while language models predict semantic meaning in text, Evo models predict the 'language of life'—genetic sequences. The research team trained the model on vast amounts of genetic code from viruses, bacteria, plants, and humans, refining it for precision so it could specifically design 'phages'—viruses that attack only targeted bacteria.

From the numerous designs generated by AI, the team pre-selected 302 optimal candidates and synthesized them artificially in the lab. Results showed that 16 of the new phages demonstrated excellent antibacterial performance, successfully eliminating E. coli in petri dishes. Samuel King, a PhD student at Stanford involved in the research, recalled that when the team observed transparent plaques forming on bacterial layers in the petri dish in the early hours of the morning—proof that the AI-designed virus was precisely consuming bacteria—the entire team erupted in applause.

For biologists and virologists, the biggest benefit of this AI advancement lies in addressing the growing problem of drug-resistant 'superbugs.' As the number and variety of such pathogens increase, rendering traditional antibiotics ineffective, this technology offers the potential for scientists to develop targeted therapies—designing phages tailored to combat specific resistant strains.

Moreover, this breakthrough signifies a new era in synthetic biology: AI is no longer limited to simulating existing natural biological structures but can now create entirely novel biological designs beyond anything found in nature. This holds immense potential for future applications in gene therapy, immunotherapy, and novel drug development.

However, this technological leap also brings dual-edged concerns. A commentary published in the prestigious international journal Science features Dr. Thomas Inglesby and Dr. Moritz Hanke from the Johns Hopkins Center for Health Security, who strongly emphasize that while generative AI’s ability to design viral genomes is now undeniable, the critical issue is ensuring the technology is not misused or causes serious harm. They particularly stress that research into any novel virus with pathogenic potential must not proceed.

To ensure safety, the Stanford team implemented strict safeguards during development: excluding data on viruses capable of infecting complex organisms from their training database, strictly limiting the scope to phages, and conducting all experiments within high-security-level laboratories.

Marc Güell, Professor of Synthetic Biology at Spain’s Pompeu Fabra University, and Patrick Cai, Professor at the Manchester Institute of Biotechnology in the UK, both agree that this study demonstrates genome language models are beginning to grasp the design principles of evolution—marking the dawn of a new era of computer-assisted genome writing.