AI creates 700,000 new viruses: Why scientists are alarmed
Scientists created artificial viruses using AI (photo: Pexels)
AI has learned to create new viruses that can infect bacteria. Scientists believe the technology could help accelerate the development of treatments for dangerous bacterial infections, according to a study published in the journal Science.
How AI learned to write DNA
Scientists used specialized genome language models Evo 1 and Evo 2.
Unlike standard chatbots, which are trained on texts and websites, these systems were trained on trillions of nucleotides—the building blocks of DNA—to learn genetic "grammar."
For the experiment, the models were additionally trained on about 15,000 viruses from the Phi X-174 family, a microscopic virus that infects only E. coli.
The researchers deliberately restricted the parameters, excluding any structures capable of infecting humans, animals, plants, or fungi.
Experiment findings:
- The Evo model generated about 700,000 potential new viruses
- Researchers selected the 285 most promising candidates
- The team created synthetic DNA for the selected variants and introduced it into bacteria
- As a result, 16 designs turned out to be fully viable viruses, and some of them reproduced even faster than the natural Phi X-174 virus.
Potential for medicine, security risks
Scientists say that AI's ability to design genomes could provide significant benefits in controlled environments.
The generated viruses could
- Improve gene therapy methods
- Expand understanding of genome structure
- Provide new tools for fighting dangerous bacteria.
At the same time, the emergence of technologies capable of designing genetic code creates significant risks in the absence of proper government regulation.
While basic chatbots can only provide information about planning biological weapons, specialized models like Evo could potentially be used to create highly contagious or deadly viruses if they fall into the hands of malicious actors.
Experts emphasize the need to urgently establish international safeguards and control mechanisms, as the pace of AI development is already outpacing current biosafety regulations.
