We look at the problem of estimating risk (Operational Risk, Credit Risk and Market Risk) and argue that risk elements, such as processes in an organization, credits in a loan-portfolio or share prices in an investment portfolio cannot be regarded as independent. This naturally leads to formulating risk models as dynamical models of interacting degrees of freedom (particles). The operational risk and credit risk problems can be cast into a language describing heterogeneous lattice gasses, in which interaction parameters and non-uniform chemical potentials have an interpretation in terms of unconditional and conditional failure probabilities. For the market risk problem, a minimal interacting generalization of the classical Geometric Brownian Motion model leads to a formulation of market dynamics that is formally similar to the dynamics of graded response neurons. We describe elements of the statistical mechanical analysis of these models to reveal their macroscopic properties.