Artificial intelligence has moved beyond generating images, answering questions and writing computer code. Scientists have now demonstrated that AI can help design complete viral genomes capable of producing functioning viruses in the laboratory.
The development could eventually open a new front in the fight against antibiotic-resistant bacteria. At the same time, it is raising serious questions about how biological AI should be controlled as the technology becomes increasingly powerful.
Researchers led by Stanford chemical engineer Brian Hie used genome-based artificial intelligence models to design bacteriophages — viruses that infect bacteria rather than humans.
Of 285 AI-generated designs tested by researchers, 16 produced functional bacteriophages capable of propagating and inhibiting the growth of targeted bacteria.
The research has attracted fresh international attention following its publication and reporting this week.
What exactly did scientists create?
The viruses involved are known as bacteriophages, often shortened to “phages”.
Unlike viruses such as influenza or COVID-19 that infect humans, bacteriophages infect bacteria.
They are naturally found throughout the environment and have been studied for decades as a possible way of treating bacterial infections.
The researchers focused on a small bacteriophage related to ΦX174, which has a genome containing only 11 genes.
Artificial intelligence models were used to generate new genetic sequences. Those sequences were then physically assembled and tested in laboratory bacteria.
Sixteen of the 285 designs successfully functioned as viruses.
Some contained genetic changes that had not been found in known natural sequences, demonstrating that the AI was capable of producing genuinely novel biological designs rather than simply copying an existing virus.
Arc Institute researchers said all 16 functional phages maintained a restricted host range and were tested using non-pathogenic laboratory strains of E. coli.
Why could this be important for medicine?
One of the biggest potential applications is antibiotic resistance.
Antibiotics have transformed medicine, but bacteria continually evolve. Some infections are becoming resistant to multiple antibiotics, making them increasingly difficult to treat.
Bacteriophage therapy offers another possible weapon.
Instead of using a chemical antibiotic to kill bacteria, doctors could potentially use viruses that specifically attack the bacteria responsible for an infection.
One problem is that bacteria can also develop resistance to bacteriophages.
Researchers therefore tested whether the diversity produced by artificial intelligence could help overcome that resistance.
According to the research team, mixtures involving AI-generated phages were able to overcome resistance in laboratory strains of E. coli where the original natural bacteriophage was unsuccessful.
That raises the possibility that future AI systems could help scientists design highly targeted phages for particular bacterial infections.
It does not mean an AI-designed phage treatment is ready to be routinely given to patients. Considerably more research and safety testing would be required before any new treatment reached widespread clinical use.
So why are biosecurity experts concerned?
The same breakthrough that makes the research medically interesting also creates an obvious dilemma.
If artificial intelligence can learn enough about genetic sequences to design a functioning viral genome, scientists and governments need to consider what increasingly capable future systems may be able to design.
Researchers involved in the work themselves acknowledged important biosafety, biocontainment and biosecurity issues.
Experts from the Johns Hopkins Center for Health Security have also warned that the capability to compose viral genomes using generative AI is developing faster than some systems designed to govern its use.
However, that does not mean researchers have created a dangerous new human virus.
The distinction is extremely important.
The successful viruses in this experiment were bacteriophages designed to infect bacteria. The researchers used non-pathogenic laboratory strains and reported extensive containment procedures.
There is currently no evidence from this work that these AI systems have successfully designed a new virus capable of infecting humans.
Researchers also deliberately implemented safeguards around the work.
Could the technology be misused in the future?
That is the question now facing governments, biotechnology companies and researchers.
Biosecurity specialists have argued that safeguards should exist at several levels rather than relying on one restriction.
Those could include controls around access to powerful biological AI models, independent research reviews, laboratory containment requirements and screening when companies manufacture synthetic DNA.
DNA synthesis screening may become particularly important because an AI-generated genetic sequence remains computer data until someone attempts to physically manufacture it.
Scientists have also pointed out that modifying existing dangerous pathogens remains considerably easier than designing an entirely new complex pathogen from scratch.
Nevertheless, the ability of AI to design even relatively simple functioning viral genomes demonstrates how quickly artificial intelligence is entering areas of science once thought extraordinarily difficult.
A breakthrough — with a difficult question attached
There is a potentially enormous positive side to the research.
If scientists can use AI to rapidly design bacteriophages targeting antibiotic-resistant bacteria, it could eventually provide doctors with new options against infections that existing medicines struggle to treat.
But advances in biotechnology frequently have what researchers call “dual-use” potential: knowledge that can be used for beneficial purposes may also present risks if deliberately misused.
That is why the debate surrounding this research should not simply be whether AI-designed viruses are “good” or “bad”.
The more important question is whether safety rules, scientific oversight and governments can keep pace with what biotechnology and artificial intelligence are becoming capable of doing.
For now, these AI-designed viruses attack bacteria — not humans.
But the breakthrough demonstrates something that only a few years ago would have sounded like science fiction:
Artificial intelligence can now help scientists design functioning viral genomes.
And that leaves society with a question it is likely to face more often as AI enters biology.
If technology gives us the ability to design entirely new forms of biology, where should the line be drawn?
Sources: Arc Institute research material, Stanford HAI and current reporting on the findings.