The prefrontal cortex (PFC) is thought to learn internal world models, but the nature of its inductive biases is unknown. We tested whether prefrontal biases are representational (selecting domain-specific features) or computational (implementing domain-general computations). During fMRI, participants learned probabilistic features of virtual environments that represented spatial, social, and sequential domain knowledge. The features were different in each domain but all domains were governed by the same mappings between features, matching their computational demands. We found no evidence for domain feature-specific representations in PFC. Instead, PFC patterns encoded a triad of specialized yet domain-general computations. Ventromedial PFC patterns resembled a Monte Carlo-like simulator, abstracting out hidden probability distributions into task states. Neural patterns in this region were sensitive to (posterior) state changes within a low-dimensional latent space. Anteromedial PFC patterns tracked local directional shifts within each task state and switches between different states organized along orthogonal axes, suggesting a global task coordinate system. Dorsomedial PFC predicted future observations, signalling task transitions. Together, these results unify diverse findings across studies, by suggesting that general-purpose world-modelling processes are implemented in medial PFC.