Your news feature outlines how designers of large language models (LLMs) struggle to stop them from hallucinating (see Nature 637, 778–780; 2025). But AI confabulations are integral to how these ...