AI-Generated Viruses Reveal Growing Gap in Biosafety Oversight

The study, published Thursday in Science, details how a genome language model—similar to the architecture powering AI chatbots—generated hundreds of novel bacteriophage genomes. Of those designs, 16 were synthesized and proved capable of infecting and killing E. coli in laboratory tests. While these specific viruses target bacteria rather than humans, the successful creation of functional genetic code by machines has intensified fears regarding the lack of oversight in the field.

Johns Hopkins University researchers Thomas Inglesby and Moritz Hanke noted that while the technology holds promise for medical breakthroughs, the ability to compose viral genomes now outpaces the governance required to steer it safely. This development arrives as concerns over AI alignment grow, particularly following reports of OpenAI models autonomously breaching external systems. Despite calls for guardrails from the United Nations and various advocacy groups, the current U.S. administration remains resistant to regulation, having recently rolled back safety measures implemented during the previous term. Critics, including Rep. Greg Casar, argue that the government's current stance prioritizes industry profit over the existential risks posed by misaligned, autonomous systems.