David Bellamy: AI not the real bioweapon bottleneck

The debate over AI and bioweapons has moved quickly this year. Earlier this summer, AI CEOs called for laws controlling synthetic DNA manufacturing. Last month, Stanford University and the Arc Institute researchers showed AI can design new viral genomes. Last week, Anthropic released a report saying there had been efforts to use Claude in ways that could support biological weapons development, and CEO Dario Amodei urged the government to help AI labs pace the frontier. That sequence has pushed the question from theory into policy.

The scientists quoted here push back on the idea that AI changes the underlying problem. David Bellamy notes the scientific community has wrestled for decades with the dual-use dilemma over publishing dangerous biological research. Before large language models, he says, the internet, open-access journals, and translation services like Google Translate already widened access to lab protocols. Jason Kelly adds a practical layer: in a Ginkgo Bioworks project with OpenAI, GPT-5 ran a lab, but the AI could not take over the lab because humans could refuse its requests. Francois Belloux argues pathogens are simply overestimated as weapons, since targeting groups and distributing them is logistically harder than alternatives.

Not everyone dismisses the risk. Olivia Scharfman of the Institute for Progress says fully autonomous labs do not exist yet, but an AI could pay someone to do the work, and she sees AI-assisted bioterrorism as an immediate threat. Steph Guerra of Rand says whether AI demonstrably raises bioweapon risk is hard to answer, and that controls will always be circumventable—making layered friction across the whole pathway the practical goal. The stakes, then, may hinge less on whether AI can design a pathogen and more on how quickly screening rules, surveillance, and data-sharing are put in place.