The demand for high-quality surface integrity and precise kerf geometry in composite machining continues to grow, especially in aerospace and automotive industries, making abrasive water jet cutting a key non-thermal machining technology. However, the process involves complex nonlinear and stochastic interactions between the abrasive jet and anisotropic composite materials, which complicate reliable prediction of kerf depth, kerf width, and surface roughness. This work aims to develop a stochastic surface evolution model for abrasive water jet cutting of anisotropic composites, integrating deterministic jet dynamics, material anisotropy, and random particle impact fluctuations. The proposed methodology is based on a stochastic partial differential equation of the Kardar-Parisi-Zhang type, extended with a spatially varying jet intensity distribution and an erosion efficiency function that depends on matrix and fiber hardness, fiber volume fraction, and fiber orientation. The radial distribution of abrasive particle energy is described by a modified Gaussian profile calibrated through Monte Carlo simulations of particle trajectories, accounting for turbulence-induced beam spreading. Numerical simulations are performed on a discretized grid with periodic boundary conditions to reproduce kerf formation and roughness development over time. The results indicate that increasing jet pressure or stand-off distance broadens the energy flux distribution, producing smoother deterministic profiles and slower kerf deepening, while low pressure and short stand-off distance generate narrow, high-intensity jets that promote stronger localized erosion and faster roughness growth. The roughness evolution follows a power-law scaling with growth exponents in the range of 0.60-0.67, indicating kinetic roughening behavior. The proposed model provides a physically consistent tool for predicting kerf geometry and surface integrity and offers practical guidance for optimizing cutting parameters to minimize surface roughness while maintaining machining efficiency.