AI's Dual Edge: The Growing Biosecurity Concern Over Designing Bioweapons

It is difficult to imagine a snail posing a fatal threat to humans, yet certain marine molluscs, known as cone snails, possess venoms capable of precisely that. Their stings deliver a complex mix of small proteins called conotoxins, some of which can disrupt the nervous system's ion channels, with no known antivenom. While many conotoxin structures are harmless or even medically beneficial – one is used in a chronic pain treatment – research into dangerous variants is highly restricted in some nations. This context explains the alarm raised in 2024 when Chinese scientists unveiled an artificial intelligence tool designed to create conotoxins. A senior US government employee, in an email seen by Nature, flagged the study as a potential biosecurity risk, particularly noting that the AI is based on an open-source protein language model developed by US scientists. However, Weiwei Xue, a co-author of the study and a computational chemist at Chongqing University, stated that the research was solely aimed at drug discovery. His team has identified potential therapeutic conotoxins through laboratory testing, emphasizing that the tool was not designed for harmful purposes and that translating AI designs into physical molecules requires significant expertise and equipment. Other researchers echoed that the immediate risks appear minimal. Nevertheless, this incident underscores a growing apprehension about emerging AI tools in biology. While these tools promise innovative drugs and societal benefits, they also present a pathway for creating new threats. The revolution in biological AI, exemplified by technologies like AlphaFold, now allows scientists to design bespoke proteins and viruses, including those that combat superbugs. General-purpose chatbots can further disseminate knowledge on laboratory design and creation. The pressing question is whether these advanced AIs could accelerate the development of more potent toxins, viruses, or other bioweapons. Interviews with over 20 scientists and policy researchers confirm the seriousness of the biosecurity threat. Martin Pacesa, a structural biologist, voiced a chilling concern: “Theoretically — and this is what keeps me up at night — one could now develop toxins on the level of ricin or other very deadly agents that would be virtually undetectable.” James Black, an AI biosecurity researcher, identifies two primary concerns: individuals using chatbots to learn about existing threats like anthrax, and more sophisticated actors (states or well-funded terrorist groups) combining chatbots with specialized biological software to engineer novel bioweapons. Researchers highlight AI-designed pandemic viruses as the greatest potential threat to humanity. This could involve modifying existing viruses, such as SARS-CoV-2 or influenza, to enhance properties like immune evasion. Existing AI tools, originally intended for surveillance and vaccine design, could be repurposed for such nefarious ends. Alternatively, AI models might design entirely new pathogens, challenging detection and countermeasures. A 2025 preprint, for instance, used AI to design bacterial viruses, with about 5% proving functional in the lab, though these were not designed to infect humans. Despite these alarming possibilities, a 2025 report by the US National Academies of Sciences, Engineering, and Medicine (NASEM) offers a crucial reality check. It concluded that numerous barriers still impede the meaningful use of AI to enhance pandemic pathogens or design them from scratch. Key obstacles include a lack of high-quality data linking virulence or transmissibility to genetic sequences, making reliable predictions difficult. Furthermore, the complexities of laboratory production and testing of pathogens remain largely unaddressed by AI. Some scientists even question the necessity for bad actors to turn to AI when the natural world already presents a multitude of inherent biological threats.