L'IA peut-elle générer de nouveaux virus avec des profils de contagiosité et de létalité prédéterminés optimisés pour l'échappement aux vaccins en utilisant des pipelines de biologie synthétique ?

Les outils d'IA simulent désormais l'évolution virale et le repliement des protéines à des vitesses sans précédent. Bien que conçus pour la préparation aux pandémies, les risques de double usage sont importants. L'IA peut-elle concevoir la prochaine pandémie ?

Background

AI systems have demonstrated substantial progress in high-throughput viral evolution modeling and high-fidelity protein folding predictions, capabilities initially developed to accelerate vaccine and therapeutic design. However, these tools operate under strict ethical and biosecurity governance frameworks and are not engineered for autonomous synthesis or deployment of novel viral pathogens. Existing AI platforms support analysis of viral genomes, such as identifying conserved epitopes or forecasting mutational trajectories in seasonal influenza or coronaviruses, but they do not autonomously generate biologically functional agents. Moreover, the physical construction of viral genomes—whether through synthetic biology pipelines, CRISPR-based editing, or cell-free expression systems—requires specialized biocontainment laboratories subject to international biosafety and biosecurity standards such as the WHO Laboratory Biosafety Manual and the Biological Weapons Convention. While computational tools can simulate protein structures and suggest mutations that might alter binding affinity or antigenicity, these outputs are hypotheses pending empirical validation under controlled experimental conditions. Current literature emphasizes proactive pandemic preparedness through surveillance and countermeasure development rather than the creation of novel pathogens, with public health authorities and scientific bodies like the National Academies of Sciences, Engineering, and Medicine underscoring the absence of validated protocols for generating optimized, harmful viral variants de novo.