CrCoNi medium-entropy alloys exhibit outstanding mechanical properties arising from pronounced chemical disorder and magnetic effects, which pose significant challenges for large-scale atomistic simulations. Here, we develop a general-purpose machine-learned interatomic potential for CrCoNi alloys within the neuroevolution potential (NEP) framework, achieving near first-principles accuracy with high computational efficiency. A comprehensive training dataset was constructed by sampling elemental, binary, and ternary configurations across relevant crystal structures and thermodynamic conditions, with all reference data obtained from spin-polarized ab initio calculations, thereby implicitly incorporating magnetic effects. The resulting NEP model accurately reproduces a broad range of key properties of elemental Cr, Co, and Ni, including equations of state, phonon dispersions, elastic constants, surface and defect energies, and ideal tensile strengths. For CrCoNi alloys, the potential reliably captures mechanical properties, stacking fault energies, and chemical short-range order over a wide range of compositions, in close agreement with density functional theory results. In addition, the model provides an accurate description of elemental interactions in the liquid state, yielding well-predicted melting temperatures. Compared with classical interatomic potentials, the present NEP model offers substantially improved accuracy while remaining more computationally efficient than previously reported machine-learning potentials, providing a robust and efficient foundation for large-scale, high-fidelity atomistic simulations of CrCoNi medium-entropy alloys.