Scientists are using AI to design novel viruses
AI-powered biotechnology is advancing faster than regulations
1First AI-designed viruses created
Researchers have successfully used artificial intelligence to design completely novel viruses. So far, only bacteriophages that infect E. coli bacteria and show 90% similarity to natural ones have been produced. However, this breakthrough raises concerns that AI could eventually design more dangerous viruses capable of infecting plants and animals. The study was published last week in the journal Science.
2Regulations lag behind
Advances in AI-powered biotechnology are rapidly outpacing existing regulations. Toby Webster from Sentinel Bio notes that boundaries are moving very quickly. In fact, the paper published in Science is already outdated; the research first appeared on bioRxiv in September 2025. The open-source model has already been fine-tuned for other applications.
3Growing biosecurity risks
Doni Bloomfield from Fordham University warns that future AI models might suggest entirely new viral sequences that evolution hasn't discovered. This could potentially enable even non-experts to design biological weapons. 'We should be extremely cautious about extending this work to viruses that can infect more complex life forms,' Bloomfield cautions.
4Current biosecurity measures
Filippa Lentzos from King's College London points out existing regulations regarding lab safety, genetic modifications, and work with dangerous pathogens. However, the real challenge lies in connecting this governance with digital biology design capabilities. As Microsoft discovered last year, current safety measures can fail to block toxic protein sequences.
5Balancing potential benefits and risks
AI-designed viruses could help combat antibiotic-resistant bacteria or create safe carriers for gene therapies. However, Bloomfield and colleagues propose a tiered access system to reduce risks related to viral transmissibility and virulence. This system would resemble current biosafety protocols that restrict access to pathogens like Ebola to high-security labs.
6Regulatory challenges
Allison Berke from RAND Center notes that knowledge gaps make regulation difficult. For instance, it's unclear whether models would need specific training data to recreate the 1918 pandemic flu virus or could extrapolate it from general datasets. 'We don't fully understand how this capability curve is progressing,' Berke states.
7Policymakers' dilemma
Tessa Alexanian from the International Biosecurity Initiative says both scientists and policymakers are discussing these issues with increasing urgency. However, U.S. lawmakers appear to be waiting for a shocking demonstration of biological capabilities before acting, likely due to fears of overregulating beneficial research. This suggests research will continue outpacing regulations in the near future.
8Call for responsible development
Webster observes that while many in the field care about responsible model development, they often lack official guidance. The current system relies largely on goodwill. Researchers are calling for more effective biosecurity approaches in their papers, but global collaboration and standards are urgently needed in this emerging field of AI-powered biological design.