A new kind of AI is now designing viruses in the lab, and the results are opening two debates at once: how to fight drug-resistant bacteria, and how to stop the same tools being misused.
Scientists in the United States used the AI model Evo 2 to create bacteriophages, viruses that can kill bacteria. Researchers at Stanford University then tested them in the lab.
In those tests, a mixture of 16 “exceptionally good” viruses designed by Evo 2 rapidly killed E. coli bacteria that were already immune to natural phages.
E. coli is a group of bacteria that can cause gut infections and is becoming increasingly resistant to available antibiotics.
The researchers said the work could help produce a new generation of antibiotics aimed at drug-resistant “superbugs”. They also noted the technology could later be used to target other harmful bacteria, including those that cause tuberculosis and MRSA, a common hospital-acquired infection.
Antibiotic resistance is rising worldwide, driven mainly by the mis- and overuse of antibiotics. That allows bacteria to develop ways to survive treatment, making existing medicines ineffective.
“If the bacteria gain resistance to a single phage, it’s game over for the medication,” said Brian Hie, chemical engineer at Stanford and co-author of the study.
“But if you have multiple genetically distinct phages in a mixture, it would be harder for the bacteria to develop resistance to the entire cocktail.”
The same work has also raised biosecurity concerns because the researchers made Evo 2 openly and freely available for anyone to download and use.
The study authors said making the tool open source has prompted safety discussions, including concern that “bad actors” could use modified versions to design harmful biological agents.
“As the authors highlight, this raises some serious regulatory and safety concerns, to say the very least,” said Simon Clarke, associate professor in cellular microbiology at the University of Reading in the United Kingdom, who did not participate in the study.
“While work of this nature is normally tightly regulated, it is reassuring that these scientists have shown further restraint in providing important guardrails, but there is no guarantee that every other scientist attempting to do something similar will be so careful,” he added.
Clarke said naturally occurring pathogens currently pose a greater risk than AI-designed ones because they are already easier to access and produce than to create from scratch.
According to the authors, one advantage of AI-designed biology over natural evolution is the ability to build safety checks directly into the process.
Evo 2 was trained on millions of natural genomes from across the world, which allowed it to learn the “grammar” and rules that make a DNA sequence functional.
Like other biology AI models, its training data includes biological information such as genetic sequences and pathogen characteristics.
There is currently no universal framework regulating these datasets. Some developers voluntarily exclude high-risk data, but researchers say clear and consistent rules should apply to all.
In February 2025, Evo 2’s team said it had excluded pathogens infecting humans and other complex organisms from its datasets because of ethical and safety risks, and to “preempt the use of Evo for the development of bioweapons”.
Earlier this year, more than 100 researchers from around the world signed an open letter saying that while open access to scientific data has sped up discovery, a small subset of new biological data carries biosecurity risks if misused.
“The stakes of biological data governance are high, as AI models could help create severe biological threats,” the authors wrote.
The researchers said the challenge is striking the right balance between openness and security restrictions on high-risk data as AI systems become more powerful and widely available.
Read more from Euro News.
