I'm sorry, you're not going to die from an AI-engineered supervirus

This item is a blog response to Noah Smith's article about AI-designed superviruses. The author says reading Smith on a topic close to their own expertise undermines their prior view of Smith as an informed commentator and cites Gell-Mann amnesia. Smith's article opens with a fictitious story about a disgruntled teenager in 2029 who asks an AI to design lethal viruses, deploys them, and kills 90% of the human population. Smith then says he has yet to hear an even halfway-convincing argument that this scenario is far-fetched. The author notes the article is paywalled, cannot read beyond the first few introductory paragraphs, and cannot post comments, so the response rests on assumptions and may be corrected if Smith addresses the points.

The author's credentials are central to the rebuttal: they say that unlike Smith, they have computationally designed viruses and other biological systems, and their day job involves developing and evaluating AI systems for protein and peptide design, giving them knowledge of state-of-the-art AI tools in biology and the challenges of designing and building functional biological systems, plus some virology. They argue computational design of biological systems is unfathomably difficult, with experts routinely failing. As an example, PhD students in 2026 using state-of-the-art AI software spend months or years trying to design simple peptide binders that inhibit an enzyme or pull down a protein, and the majority of designs fail, do not express, or are toxic. In Smith's fictional world, an untrained teenager solves a problem thousands of times more complicated than designing a peptide binder, and the distance between today and that level of capability is enormous.

Even if someone could design the perfect virus, the author says assembling and distributing it would be a non-trivial task in its own right; one cannot simply order a working virus from temu.com, and they cite an article by Abi Olvera. The author does not rule out that somebody may eventually cause harm with an AI-designed biological agent, but argues that scale must be considered: a disgruntled teenager could also get their hands on a fertilizer bomb.