While a significant body of investigations have been focused on the process of protein self-assembly, much less is understood about the reverse process of a filament breaking due to thermal motion into smaller fragments, or depolymerization of subunits from the filament ends. Indirect evidence for actin and amyloid filament fragmentation has been reported, although the phenomenon has never been directly observed either experimentally or in simulations. Here we report the direct observation of filament depolymerization and breakup in a minimal, calibrated model of coarse-grained molecular simulation. We quantify the orders of magnitude by which the depolymerization rate from the filament ends $k_mathrm{off}$ is larger than fragmentation rate $k_{-}$ and establish the law $k_mathrm{off}/k_- = exp [( varepsilon_| - varepsilon_bot) / k_mathrm{B}T ] = exp [0.5 varepsilon / k_mathrm{B}T ]$, which accounts for the topology and energy of bonds holding the filament together. This mechanism and the order-of-magnitude predictions are well supported by direct experimental measurements of depolymerization of insulin amyloid filaments.