The article examines whether AI could make dangerous biological work substantially easier, rather than focusing only on whether it can independently invent a civilization-ending virus. It begins with a Science paper showing that genome language models can help design functional bacteriophages, viruses that infect bacteria. Responses to that work have emphasized the large gaps between generating a biological design and turning it into a working, dangerous agent, including the need for expertise, equipment, experiments, and practical experience. The author accepts those barriers but argues that they do not settle the broader question of whether AI can make existing biological knowledge more usable to smaller groups. The article points to anthrax and the 1979 accidental release at a Soviet biological facility in Sverdlovsk as reminders that existing pathogens can cause severe harm without AI. It also describes historical Soviet efforts to make bacteria resistant to multiple antibiotics and notes that Soviet programs pursued contagious-disease weapons, while cautioning that this history does not prove they successfully increased human-to-human transmission. Medical treatment is not a complete safeguard: a large outbreak could overwhelm hospitals, supplies, distribution systems, and emergency services even when treatment exists. The author says AI may reduce some expertise barriers by explaining unfamiliar technical knowledge, but stresses that convincing answers can be wrong and experiments still fail. Evidence is still needed to show how much AI changes what people can accomplish in real biological work. The concern is therefore not that a small group can immediately reproduce a massive state program, but that even a partial reduction in missing expertise could matter to malicious actors. The article concludes that uncertainty about AI-designed superviruses should not distract from measuring whether AI lowers the practical barriers to using known biological capabilities.
