Innovative materials designed to endure the extreme conditions found in fusion devices are frequently investigated through molecular dynamics (MD) simulations, which allow researchers to examine material characteristics and behaviors under relevant loading scenarios. To enhance computational efficiency, these MD simulations typically utilize interatomic potentials (IAPs) that are calibrated to align with data from more resource-intensive, high-fidelity methods such as density functional theory. However, this calibration introduces a degree of uncertainty. This study focuses on quantifying the uncertainty linked to two specific IAPs: the Spectral Neighbor Analysis Potential (SNAP) and the Atomic Cluster Expansion (ACE). The uncertainty for SNAP is assessed using an embedded model error framework, where uncertainty is integrated into the model by assigning a distribution to the model coefficients. The mean and covariance of this distribution are then found within an inverse modeling framework, incorporating model form into the uncertainty assessments of derived energies, forces, and stresses across various materials. The embedded model error method is contrasted with conventional Bayesian linear regression. Additionally, the development of a polynomial chaos surrogate is investigated to facilitate more efficient forward propagation of uncertainties toward estimation of elastic properties, binding energies, and defect characteristics derived from the ACE potential.