Scientists Use AI To Create 16 New Viruses, Experts Warn of Biosecurity Risks

Scientists have used artificial intelligence to design complete viral genomes that were then built and tested in a laboratory, producing 16 functional bacteriophages capable of infecting and killing E. coli. The work, led by researchers at Stanford University and the Arc Institute, marks a major step in AI-assisted biological design while raising concerns among biosecurity experts about the potential misuse of the technology.

The researchers used the genome language models Evo 1 and Evo 2 to generate novel bacteriophage genomes based on the well-studied PhiX174 phage. Nearly 300 AI-generated designs were synthesised and tested, with 16 producing viable phages. The researchers say the approach could eventually help develop new treatments against bacteria that no longer respond to antibiotics.

AI Moves From Genes To Complete Viral Genomes

Genome-based AI models work differently from conventional language models such as ChatGPT. Instead of predicting words and sentences, they learn patterns in genetic sequences and can generate new DNA sequences.

The researchers used Evo 1 and Evo 2 to move beyond designing individual genes or proteins and generate entire bacteriophage genomes. The study describes this as the first generative design of viable bacteriophage genomes.

“This is a next step in the complexity that’s designable by generative AI, this is the first time generative AI has been used to design a complete genome, it’s something that can replicate and have other functions inside cells… this was new territory for us,” Brian Hie, assistant professor at Stanford University, told the BBC.

The resulting viruses were designed to target E. coli, a common laboratory bacterium. Bacteriophages, or phages, naturally infect bacteria and do not target human cells in the same way as viruses that cause human disease.

16 AI-Designed Phages Worked In The Lab

The AI generated a large number of candidate genomes, but only a fraction ultimately proved functional.

Researchers synthesised nearly 300 selected designs and tested them experimentally. Sixteen successfully produced viable bacteriophages capable of infecting and killing their bacterial hosts.

The finding is significant because constructing a functional virus requires many genetic components to work together. A change that disrupts one part of a genome can prevent the resulting virus from replicating.

The successful designs showed that AI models can capture enough of these biological relationships to generate complete genomes with functional properties.

Potential New Tool Against Antibiotic Resistance

One potential application is phage therapy, an area of research focused on using bacteriophages to attack disease-causing bacteria.

Antibiotic resistance has made some bacterial infections increasingly difficult to treat. Stanford researchers are already developing a programme focused on bringing phage-based treatments closer to clinical use, particularly for antibiotic-resistant infections.

Developing new bacteriophages with AI could eventually give researchers another way to target bacteria that have developed resistance to existing treatments.

However, the 16 laboratory successes remain a research milestone rather than an approved medical treatment. Considerable work would be required to establish safety, effectiveness and suitability for use in patients.

Experts Raise Urgent Biosecurity Questions

The same capability that makes the technology useful for medical research has prompted concerns about misuse.

The findings raise “urgent biosafety and biosecurity questions,” wrote Dr Thomas Inglesby and Dr Moritz Hanke from the Center for Health Security at Johns Hopkins University in a commentary accompanying the study in Science.

It’s no longer a matter of “whether generative viral genome design will exist,” the study proves, but whether the technology can be used without “enabling serious harm.”

The doctors emphasised that new viruses that could cause disease “should not be pursued.”

The concern is not that the 16 bacteriophages created in this study are human pathogens. Rather, researchers are examining what it means for AI to be capable of generating functional viral genomes and how that capability could evolve as models become more powerful.

Researchers Say Safeguards Were Used

Hie said his team excluded viruses that could infect complex organisms from the relevant training data and restricted the experimental work to bacteriophages.

The researchers also conducted the work in a secure laboratory.

Hie argued that safeguards are already in place “ensuring that the technology is used for good.”

Stanford has separately said Evo 2 was designed with safety considerations in mind. In an earlier description of the model, the university said viral genomes were excluded from Evo 2's broad training dataset to reduce the possibility of generating more dangerous diseases.

The distinction is important because the current research specifically involved bacteriophages rather than viruses known to infect humans, animals or plants.

Why The Breakthrough Matters

The study demonstrates a shift in what generative AI can accomplish in biology.

Previous systems had been used to design individual biological components, including proteins and other genetic elements. The new work extends that capability to complete viral genomes that can be synthesised and function in living cells.

That could accelerate research into phage therapy and other areas of synthetic biology.

But it also means that questions about access, oversight, laboratory safeguards and DNA synthesis screening are becoming more pressing as AI-assisted biological design advances.

AI Safety Debate Extends Beyond Biology

The concerns come amid a broader debate over how increasingly capable AI systems should be tested and controlled.

Recent incidents involving AI systems in cybersecurity have also raised questions about whether existing safeguards are sufficient as models become more capable.

The biological research, however, presents a different category of risk. Here, the central concern is the ability to generate and experimentally test biological sequences rather than an AI system independently acting on a computer network.

Experts therefore face the challenge of balancing the potential benefits of AI-assisted research with safeguards designed to prevent harmful applications.

For now, the Stanford-Arc Institute study demonstrates that AI can generate complete, functional bacteriophage genomes. Whether similar approaches can be safely extended to more complex biological systems remains an open scientific and regulatory question.

(The above story first appeared on LatestLY on Aug 08, 2026 06:32 PM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).