A groundbreaking AI system called Delphi-2M can now predict a person’s risk of developing more than 1,000 diseases—some up to 20 years in advance—using only routine health data. Developed by researchers including Moritz Gerstung from the German Cancer Research Center, the tool is built on a modified version of a large language model (LLM), similar in architecture to chatbots like ChatGPT. Unlike most existing AI health tools that focus on a single disease, Delphi-2M analyzes a broad spectrum of conditions, including cancers, skin disorders, and immune-related illnesses, by learning from patterns in 400,000 individuals’ health records from the UK Biobank. The system takes into account age, sex, body mass index, lifestyle habits such as smoking and alcohol use, and medical history to generate personalized, long-term disease risk forecasts. Remarkably, it outperformed traditional single-disease prediction models and even biomarker-based systems in accuracy for many conditions. According to Gerstung, the model’s ability to simulate an entire future health trajectory is “astonishing,” offering a holistic view of potential health outcomes that could help clinicians identify high-risk patients early and implement preventive strategies. Experts like Stefan Feuerriegel from Ludwig Maximilian University of Munich praise the innovation, noting that the ability to model multiple diseases simultaneously could revolutionize preventive medicine. The tool’s success hinges on its capacity to learn complex, non-linear relationships between health data and disease progression—something traditional statistical models struggle with. However, the model’s training is limited to data from a single population in the UK, raising questions about its generalizability to other ethnicities, healthcare systems, and environments. Researchers caution that further validation across diverse populations is essential before clinical use. In other science news, a new study reveals that refugees hosted by local families integrate more successfully into their adoptive countries than those placed in institutional settings. The findings suggest that personal, community-based support fosters stronger social ties, language acquisition, and employment outcomes. Meanwhile, engineers have developed a “jelly-filled” garment that uses a gel-like material to keep wearers cool in hot, humid conditions—offering a promising solution for climate-resilient clothing. On the ethics front, a new study shows that people are more likely to cheat on tasks—such as tax reporting—when they can delegate them to an AI. Researchers found that the mere presence of an AI tool increases dishonest behavior, raising concerns about how automation may erode personal accountability. In technology, Europe has launched JUPITER, one of the world’s most energy-efficient AI supercomputers, aimed at boosting the continent’s AI research and innovation. At the same time, scientists have created a detailed catalogue of octopus arm movements, helping decode the complex behavior of these intelligent invertebrates—offering insights into neural control and potential applications in robotics. The Nature Podcast episode also touches on broader themes, including how AI is reshaping science and society, and why evaluating AI’s societal impact must begin now. With tools like Delphi-2M, AI is poised to transform healthcare, but challenges around fairness, transparency, and ethics remain critical. As AI continues to advance, balancing innovation with responsibility will be key to ensuring its benefits are realized safely and equitably.