Stanford And Arc Institute Create First AI-Designed Viruses Capable Of Infecting Bacteria

In Brief

Researchers at Stanford University and the Arc Institute used an AI model to design entirely new viruses that can infect bacteria, carrying hundreds of novel mutations and overcoming natural bacterial defenses.

Researchers at Stanford University and the nonprofit organisation Arc Institute have achieved a major milestone in computational biology by creating the first entirely AI-designed viruses capable of infecting and killing bacteria

The team trained an advanced AI model named Evo, which works on the same principles as large language models (LLMs) like ChatGPT, on a dataset of two million viral genomes, equipping it with the ability to understand viral structure, gene interactions, and functional constraints.

Using this model, scientists tasked Evo with designing completely new viruses from scratch, resulting in 302 unique designs, of which 16 were confirmed to be functional in laboratory tests, demonstrating the AI’s capacity to produce viable biological systems that had never existed before.

These AI-generated viruses carried up to 392 mutations that have never been observed in nature, including combinations of genetic elements that researchers had previously tried and failed to assemble using conventional engineering techniques.

Interestingly, when bacteria evolved resistance to natural viruses, the AI-designed viruses were able to overcome these defenses within days, whereas traditional viral counterparts were rendered ineffective

One particularly notable synthetic virus successfully incorporated a key protein component from a distantly related virus, an achievement that had eluded scientists for years despite repeated attempts using standard genetic engineering approaches

AI-Designed Viruses Mark A New Era In Genome Engineering

The development of these AI-designed viruses signals the dawn of a new era in scientific research, where computational tools can move beyond reading and writing genomes to actively designing them. As the Arc Institute highlighted, “this represents a new chapter in our ability to engineer biology at its foundational level.”

Researchers emphasize that their AI was intentionally not trained on viruses that infect humans. Nevertheless, this technology carries inherent risks, as it could potentially be applied by others—whether out of curiosity, scientific interest, or malicious intent—to explore human pathogens and create novel levels of virulence

Furthermore, the ability of AI to generate complete genomes for more complex organisms remains uncertain, and there is currently no straightforward method to test such designs. Unlike some viruses that can be activated directly from a DNA strand, larger organisms such as bacteria, mammals, or humans require incremental genetic modifications of existing cells, a process that remains slow and labor-intensive

Despite these challenges, scientists argue that pursuing this line of research is important. They envision automated laboratories in which AI-generated genomes could be proposed, experimentally tested, and refined through iterative feedback, potentially accelerating the development and understanding of complex biological systems.