Graphite capsules are commonly used in high-temperature, high-pressure experiments, particularly for nominally anhydrous experiments and iron-bearing silicate samples. Due to the presence of graphite in the sample assembly, the oxygen fugacity for these experiments is thought to be relatively low, typically at or below the graphite-CO-CO2 buffer (CCO). The detailed mechanism and kinetics of redox control in graphite capsule experiments are, however, poorly understood. This is especially problematic for short duration experiments (e.g. kinetic experiments), because it is uncertain whether the experimental product will preserve its initial oxygen fugacity, or become reduced during the experiment. In this study, a set of basaltic glasses after high-temperature experiments in graphite capsules were analyzed by micro X-ray absorption near-edge structure (µ-XANES) to obtain their Fe3+/ΣFe profiles near the graphite–melt interface. The results show rapid reduction of ferric iron in the basaltic melt, reaching near-equilibrium in half an hour for samples of 2 mm diameter and 1.3–1.9 mm thickness. Even for a “time-zero” experiment, which was quenched immediately after reaching the target temperature, the reduction profile is over 100 µm in length. By comparing experiments at the same temperature and pressure but with different durations, the reduction reaction progress is found to be linear to the square root of duration, indicating that the reduction process is diffusion-controlled. Such a rapid reduction of the basaltic melt requires a mechanism that is significantly faster than divalent cation diffusion or oxygen diffusion, and is best explained by molecular hydrogen diffusion. It has been shown by previous studies that nominally anhydrous high-pressure experiments could contain significant amounts of water. Thousands of ppm of H2O could remain in the graphite capsule even after drying at 120 °C for an extended time period. At high temperatures, H2O reacts with graphite to produce molecular hydrogen, which then diffuses into the basaltic glass and causes reduction. This mechanism is also supported by a compensating H2O profile of equivalent length in the basaltic glass, showing evidence for H2O produced by molecular hydrogen reacting with ferric iron. A quantitative model is proposed and it successfully reproduces the Fe3+/ΣFe profiles in our experiments. The model helps explain the kinetics of the reduction process and demonstrates that for a basaltic glass with reasonable initial FeO* content, Fe3+/ΣFe ratios, and thicknesses, the equilibrium oxidation state can usually be reached in one hour at ~ 1300 °C and ~ 0.5 GPa. Although extrapolating our conclusion to the large range of graphite capsule experiments requires knowledge on how H2 solubility and diffusivity varies as a function of silicate composition, temperature, and pressure, the reduction process is expected to be rapid in general because H2 diffusivity is high in silicate melts. Our study elucidates the mechanism and rate of oxygen fugacity change in graphite capsule experiments. Based on thermodynamic calculations, the reaction between graphite capsule and H2O is expected to produce a C-O-H fluid with an intrinsic oxygen fugacity of CCO −0.9, which agrees well with the measured Fe3+/ΣFe ratios in the basaltic glasses and the estimated oxygen fugacity for graphite and Pt–graphite capsule experiments from a previous study. Future studies are needed to better constrain the kinetics of this process at different temperature, pressure, and in silicate melts of different compositions. The dynamic process of H2 diffusing and reducing Fe3+ to Fe2+ shown in our experiments also provides a potential way to determine the diffusivity and solubility of molecular hydrogen in silicate melts, which are crucial in understanding volatile behaviors on reducing planetary bodies, such as the Moon and Mercury.