AI Creates New Viruses? Not Quite. AI-Designed Bacteriophages

Novel AI-designed viruses are now closer to being synthesised. A new Science study shows that genome language models can generate bacteriophages, creating a potential new route to antibacterial therapies while raising difficult questions about how biological AI should be governed.

Image

From DNA Prediction to Functional Phages

The latest cutting-edge research, led by scientists from Stanford University and the Arc Institute, used genome language models called Evo to explore whether AI could generate complete bacteriophage genomes with functional properties.

Bacteriophages, or phages, are viruses that infect bacteria rather than human cells.

The researchers focused on a well-characterised phage family and used AI to generate novel genome designs. They then tested selected candidates experimentally (rather than simply stopping at the computational predictions and relying on those as evidence of biological function).

The result?

Sixteen novel AI-generated phages were proven experimentally functional. Several showed useful biological characteristics, including activity against bacterial strains and, in some experiments, performance that exceeded the natural reference phage.

AI didn’t just identify an existing therapeutic candidate. It helped generate biological designs that did not previously exist in nature.

AI-Designed Bacteriophages Could Help Defeat Antimicrobial Resistance

The timing of this research into AI-designed bacteriophages is particularly relevant as antimicrobial resistance continues to challenge conventional anti-infective development.

Phages offer a completely different approach by naturally targeting bacteria. They can also be highly selective, potentially allowing therapies to attack pathogenic organisms while limiting disruption of beneficial microbial communities.

Stanford’s new Center for Phage Pharmaceuticals is also pursuing clinical applications of phage therapy against drug-resistant infections. 

AI could eventually expand the search space beyond phages that researchers can find and characterise in nature.

This is important in precision antibacterial therapy, where the challenge is not simply finding something that kills bacteria, but identifying an agent that can address a particular bacterial target while retaining the required specificity.

Pharma is already exploring AI across drug discovery and development. Pharmatica's analysis of AI R&D investment examines why scientific capability alone is not enough. Technologies must also demonstrate workflow value, scientific credibility, and a route into regulated environments.

Read Reimagine AI-driven Drug Discovery with Pharma Superintelligence

Reimagine AI-driven Drug Discovery with Pharma Superintelligence

Explore how AI-driven drug discovery is evolving through Pharmaceutical Superintelligence and what your organisation can do to keep up.

AI-Designed Bacteriophages Raise the Biosecurity Stakes, Again

The therapeutic potential comes with a difficult second question of what happens when AI can generate functional biological systems rather than merely analyse them?

The researchers deliberately built safety considerations into this work. The Evo 2 development programme excluded eukaryotic viruses from training, and the bacteriophage work exclusively focused on viruses that infect bacteria.

The objective was to demonstrate genome-scale generation while reducing the possibility of producing viruses capable of infecting humans. 

Some biosecurity experts nevertheless see the development as a warning that biological AI capabilities are advancing faster than governance frameworks.

The concern is broader than this particular experiment. There is concern about the future ability of increasingly capable models to generate biological designs that move beyond what researchers have already observed.

The safety challenge therefore shifts from prediction to generation.

Existing oversight mechanisms were largely developed around physical biological materials, laboratory practices, and known sequences. Generative models introduce another layer of risk, this time, in the digital design process itself.

That makes AI governance increasingly relevant to R&D strategy rather than simply IT policy.

Image

What Pharma Should Watch for Biosecurity

It is still unclear whether this is a niche research capability or the beginning of a broader design paradigm.

Three developments deserve attention:

• Therapeutic validation: AI-generated phages will need rigorous testing for specificity, efficacy, safety, stability, and manufacturing feasibility before they can become medicines.
• Regulatory credibility: AI-generated biological designs will require transparent evidence showing how models were used and how their outputs were validated.
• Biosecurity controls: Organisations will need governance spanning model access, biological synthesis, screening, laboratory validation, and downstream use.

The U.S. Food and Drug Administration (FDA) already takes a risk-based approach to AI used in drug and biological product development. Its guidance emphasises defining a model's context of use and establishing appropriate evidence for model credibility. 

That framework was not designed specifically for generative viral genome design. But its emphasis on risk, context, validation, documentation, and human oversight provides an important foundation.

Pharmatica has also examined the FDA’s expanding acceptance of new approach methodologies in preclinical drug development, with clear indication that computational and biological technologies are moving closer to the core of regulated R&D.

A New Frontier of Biological AI?

AI-designed bacteriophages demonstrate something larger than a new way to find phage candidates. They show that genome foundation models can more easily move from learning biological patterns to generating biological designs that can be tested in the real world.

That creates an opportunity to rethink how it approaches antibacterial discovery, particularly where conventional approaches struggle against rapidly evolving pathogens.

It also creates a governance problem that cannot be separated from innovation. As biological AI becomes more capable, scientific validation, regulatory strategy, and biosecurity will need to develop together.

At Pharmatica, we track the technologies reshaping Pharmaceutical R&D, from AI and computational biology to new therapeutic platforms and regulatory innovation. Our Insights connect emerging science with the strategic decisions pharma needs to make as the boundaries of drug development change.

Pharmatica: Insight. Connection. Impact.

Image