Knowledge graphs encode reality with crisp assertions, while the systems that now consume them—foundation models and autonomous agents—reason in probabilities. This mismatch explains why the integration of language models and knowledge graphs remains a data‑feeding pipeline rather than a unified reasoning architecture.\ \ We propose Semantic Bayesian World Models (SBWMs), a web that describes the world not as a static fact database but as a shared, evolving fabric of beliefs. Ontological axioms constrain prior distributions, observations update beliefs via Bayesian conditioning, and actions intervene on the world.\ \ An agent equipped with an SBWM gains several capabilities: a home‑security agent can infer whether a figure at the gate is a courier or a burglar; actuarial estimates can be aggregated through entailment instead of raw string frequency; planning tasks that typically cause language models to fail become tractable; and quantities never explicitly stated in any document can be estimated.\ \ Realizing this vision requires the community to build: belief annotation over RDF 1.2, probabilistic entailment regimes, semantic calibration layers, and protocols that allow never‑met agents to exchange and disagree over calibrated beliefs.\ \ Review