Why AI-Engineered Superviruses Remain Far-Fetched

Summary

In this opinion essay, computational biologist Claus Wilke challenges the idea that a non-expert could soon use AI to design a virus capable of killing most of humanity. Drawing on experience with computationally designed viruses and AI systems for protein and peptide design, he argues that even relatively simple biological designs frequently fail to function, express, or avoid toxicity. Designing a virus is only one part of the problem: synthesis, assembly, and distribution would create additional barriers. Wilke also argues that biology imposes difficult tradeoffs. Highly lethal viruses often spread poorly because sick patients have limited contact with others, while highly transmissible respiratory viruses are often less lethal. Long asymptomatic incubation combined with efficient transmission and severe later disease is not a demonstrated viral design capability. The essay further says that total harm depends on human behavior: very lethal diseases may prompt isolation, while moderately severe diseases can spread widely, as COVID did. Wilke does not claim that AI-related biological harm is impossible, but says a civilization-ending engineered virus is highly implausible for at least the coming decades. He argues that natural outbreaks, including measles, Ebola, and HIV, deserve greater attention, alongside surveillance, vaccination, vaccine research, antivirals, indoor-air improvements, and other public-health measures.