GoKawiilLarge language models (LLMs), such as those behind the chatbot ChatGPT, are increasingly used to perform actions in the real world, from sending e-mails to executing financial transactions. As the capabilities of artificial-intelligence systems grow, the technology has the potential to create valuable tools, but also to pose catastrophic risks. Writing in Nature, Cloud et al.1 report that training LLMs on AI-generated data, which is becoming increasingly common as model developers reach the limits of freely published, human-generated content, can transmit undesirable traits from one model to another. This can occur even with a rigorous screening process that excludes directly malicious content.
Nature 652, 574-575 (2026)
doi: https://doi.org/10.1038/d41586-026-00906-0
References Cloud, A. et al. Nature 652, 615–621 (2026). Betley, J. et al. Nature 649, 584–589 (2026). MacDiarmid, M. et al. Preprint at arXiv https://doi.org/10.48550/arXiv.2511.18397 (2025). Fang, L. et al. Preprint at arXiv https://doi.org/10.48550/arXiv.2504.14772 (2026). Bai, Y. et al. Preprint at arXiv https://doi.org/10.48550/2204.05862 (2022). Download references
Competing Interests The authors declare no competing interests.
Related Articles
Subjects