Toward relational biosecurity: understanding AI-enabled biology as a connected system

Le résumé fourni par la source

Advances in artificial intelligence (AI) are rapidly expanding what can be predicted and produced in biotechnology, transforming the life sciences into a computational, distributed, and increasingly design-oriented enterprise. At the same time, biosecurity frameworks remain largely organized around the control of known agents, materials, and individual technologies, creating a growing mismatch between emerging capability and current approaches to risk. In AI-enabled biology, capability is increasingly compositional, arising from interactions among data, models, infrastructure, and workflows. Risk no longer resides solely within individual components, but in how they are connected. Safeguards that are effective in isolation may fail when systems are integrated; for example, nucleic acid sequence screening may not capture risks arising from generative systems that explore novel biological space beyond reference frameworks. This Perspective argues for a relational approach to biosecurity that treats interactions between components as explicit objects of design and governance, rather than implicit byproducts of integration. It emphasizes system-level sensing, preservation of context and uncertainty, buffering of perturbations, and alignment with shared values across distributed actors. Rather than replacing existing safeguards, this approach extends them by addressing how they interact within workflows. It introduces the need to represent, measure, and propagate system-level objectives across components, enabling more coherent oversight of compositional workflows. While grounded in biological applications, these challenges extend to AI safety and governance more broadly, where risks emerge from interactions among systems rather than individual models.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.