AI simulations of liquid flow create beautiful, realistic patterns that actually break the most basic laws of physics.
April 23, 2026
Original Paper
AI models of unstable flow exhibit hallucination
arXiv · 2604.20372
The Takeaway
Neural networks are frequently used to predict how air or water moves around objects like wings or hulls. These models often produce hallucinations that look visually perfect to a human eye but violate the conservation of mass. A simulation might show a stunning swirl of smoke that literally creates matter out of nothing. This discovery warns that visual plausibility is a dangerous metric for scientific accuracy in machine learning. Relying on these unverified visuals could lead to catastrophic failures in engineering or climate modeling.
From the abstract
We report the first systematic evidence of hallucination in AI models of fluid dynamics, demonstrated in the canonical problem of hydrodynamically unstable transport known as viscous fingering. AI-based modeling of flow with instabilities remains challenging because rapidly evolving, multiscale fingering patterns are difficult to resolve accurately. We identify solutions that appear visually realistic yet are physically implausible, analogous to hallucinations in large language models. These hal