World models have evolved from compact latent dynamics to generative, controllable, and interactive simulators of embodied environments. High predictive likelihood and visual fidelity do not automatically ensure that a model retains the evidence needed for safe decision‑making. Current approaches exhibit three structural mismatches: likelihood versus risk, prediction versus intervention, and finite‑horizon prediction versus accumulated consequences. To address this, we propose the Risk‑Informed World Model (RIWM), which organizes modeling around consequences, intervention, epistemic uncertainty, and recoverability, and integrates four interdependent capabilities: decision‑relevant representation, counterfactual reasoning, safety‑critical episodic memory, and runtime safety assurance. RIWM distinguishes physical, social, and operational consequences while using epistemic uncertainty to qualify the evidence supporting actions. We further outline open challenges: identifying consequential futures, validating counterfactual reasoning, maintaining revisable safety memories, translating learned consequences into executable constraints, and determining when evidence is sufficient to act. We argue that future world models should move beyond predicting the most likely futures toward identifying which futures matter, revising judgments through experience, and appropriately choosing to act, revise, sense, defer, or abstain.
Review: RIWM embeds safety considerations directly into world modeling, offering a coherent framework that links evidence to action for safety‑critical embodied systems and inviting rigorous empirical study.