AI can design viruses, toxins and other bioweapons. How worried should we be? - Nature

A senior US government employee has flagged a study on AI-designed conotoxins as a potential biosecurity risk. The study, published in Nature, describes the design of novel conotoxins using an open-source protein language model. Conotoxins are neurotoxic peptides originally derived from cone snail venom. The employee expressed concern that AI could accelerate the development of novel bioweapons.

The AI model used is based on a publicly available protein language model, raising questions about dual-use risks. The study itself was conducted by researchers from an unnamed institution, and the Nature article notes that the work passed institutional biosafety review. The government employee flagged the study internally, not through public channels, highlighting ongoing discussions about AI biosecurity governance.

The article contextualises this within a broader debate about AI and bioweapons. Previous studies have shown AI can design known toxins and even suggest novel molecules with potential toxicity. The Nature piece concludes that expert consensus varies on the level of immediate risk.