Rana, S., Singh, H. (2026). Evolvable AI needs operational risk thresholds. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2618781123