Stanford's Evo 2 AI designs new phages that attack E. coli

Scientists at Stanford University have used the Evo 2 artificial intelligence model to design brand new viruses that can attack E. coli bacteria. The phages, which are viruses that infect bacteria, were generated entirely by AI and do not exist in nature.

The breakthrough, reported by multiple outlets including the BBC and The New York Times, marks a significant step in the use of large genome models for biological design. But it has also triggered fresh concerns about the safety and ethics of AI-created viruses.

How Evo 2 works

Evo 2 is a large language model trained on vast datasets of genetic sequences. Much like ChatGPT generates text, Evo 2 generates DNA sequences. The Stanford team fed the model with phage genomes and asked it to produce new ones that could infect E. coli.

The resulting phages were not copies of existing viruses. They were novel designs, with sequences that had never been observed in nature. When tested in the lab, the AI-generated phages successfully targeted and killed E. coli bacteria.

Medical promise and peril

The applications are potentially huge. Phage therapy, the use of viruses to treat bacterial infections, has been around for over a century but has been limited by the difficulty of finding the right phage for each bacterium. AI could accelerate that search dramatically.

With antibiotic resistance rising globally, new tools to fight bacterial infections are urgently needed. AI-designed phages could offer a tailored, rapid response to drug-resistant strains.

But the same technology raises alarm bells. Security experts and ethicists have pointed out that AI capable of designing viruses could be misused to create pathogens with harmful properties. The Guardian has reported safety fears among scientists about the first viruses designed by AI.

Safety concerns in focus

No dangerous agents were created in this instance. The phages produced are specific to bacteria and do not infect human cells. However, the principle is what worries observers: if AI can design a virus to attack E. coli, it can also design one to attack human cells.

Stanford researchers have been careful to emphasize that their work is aimed at beneficial applications. They have also called for robust oversight of AI models with biological capabilities.

Regulators and policymakers are now grappling with how to govern such tools. The debate is moving from theory to practice, as AI models become more powerful and more accessible.

What happens next

The Stanford team plans to refine Evo 2's phage designs, testing them in animal models and eventually in clinical settings. They also intend to publish the methods openly, in the spirit of scientific collaboration.

But open publication is itself a point of contention. Some argue that releasing the code and methods could enable bad actors. Others say that secrecy would hamper legitimate research and global health preparedness.

The coming months will likely see more conversations between AI developers, biologists, and security agencies. The balance between innovation and safety has never been more delicate.

For now, the Evo 2 phages remain a laboratory curiosity with real potential. Whether they become a medical tool or a cautionary tale depends on how the technology is managed.