AI Is Now Designing Viruses From Scratch. Experts Say The Rules Aren’t Ready
Scientists have used artificial intelligence to create functional viruses that have never existed in nature, a scientific milestone that could accelerate treatments for antibiotic-resistant infections while raising fresh concerns over how governments regulate increasingly powerful AI tools used in biotechnology.
The research, led by scientists at Stanford University and published Thursday in the journal Science, used a genome language model to generate synthetic bacteriophages, viruses that infect bacteria rather than humans or animals. After producing hundreds of AI-generated genome designs, researchers synthesized the most promising candidates in the laboratory and found that 16 successfully infected and destroyed Escherichia coli (E. coli) bacteria.
The achievement marks the first time generative AI has been used to design complete viral genomes that function as living biological systems, representing a significant advance for synthetic biology. Researchers said the work could eventually help develop new phage therapies against antibiotic-resistant bacterial infections, an area of growing interest as antimicrobial resistance continues to threaten global health, Financial Times reported.
The team trained its AI model on millions of bacteriophage genomes while excluding viruses that infect humans, animals or plants as part of efforts to reduce potential risks. The researchers quoted by Wall Street Journal said the study was intended as a proof of concept demonstrating that AI can generate entirely new functional genomes rather than simply modifying existing organisms.
Despite the medical promise, the findings have renewed debate over biosecurity as AI systems become increasingly capable of designing biological molecules and organisms.
In an accompanying commentary published in Science, biosecurity experts Thomas Inglesby and Moritz Hanke of Johns Hopkins University said the technological capability has now outpaced governance frameworks designed to oversee it. They wrote that while the Stanford researchers adopted appropriate safeguards, broader oversight for AI-generated genomes remains inadequate.
Experts say the immediate public health risk remains limited because the viruses created in the study infect only bacteria and are far less complex than viruses that infect humans. However, they warn that as AI models improve, similar techniques could eventually be applied to more sophisticated organisms, making it easier to design biological systems faster than regulators can respond.
According to The Guardian, the breakthrough also highlights a growing policy gap. Last month, the Trump administration introduced new rules aimed at restricting federally funded high-risk life sciences research, including certain gain-of-function experiments involving dangerous pathogens. Those policies, however, do not specifically address AI-generated biological designs, a rapidly evolving area that many scientists believe requires dedicated oversight.
Researchers and biosecurity specialists increasingly argue that safeguards should extend beyond AI models themselves to include DNA synthesis companies, laboratory biosafety practices and screening systems capable of detecting potentially dangerous genetic sequences before they are manufactured. Such a layered approach, they say, would better address future risks as AI becomes more deeply integrated into biological research.
