Large-scale underground hydrogen storage (UHS) plays a vital role in energy transition. H2-brine interfacial tension (IFT) is a crucial parameter in structural trapping in underground geological locations and gas-water two-phase flow in subsurface porous media. On the other hand, cushion gas, such as CO2, is often co-injected with H2 to retain reservoir pressure. Therefore, it is imperative to accurately predict the (H2 + CO2)-water/brine IFT under UHS conditions. While there have been a number of experimental measurements on H2-water/brine and (H2 + CO2)-water/brine IFT, an accurate and efficient (H2 + CO2)-water/brine IFT model under UHS conditions is still lacking. In this work, we use molecular dynamics (MD) simulations to generate an extensive (H2 + CO2)-water/brine IFT databank (840 data points) over a wide range of temperature (from 298 to 373 K), pressure (from 50 to 400 bar), gas composition, and brine salinity (up to 3.15 mol/kg) for typical UHS conditions, which is used to develop an accurate and efficient machine learning (ML)-based IFT equation. Our ML-based IFT equation is validated by comparing to available experimental data and other IFT equations for various systems (H2-brine/water, CO2-brine/water, and (H2 + CO2)-brine/water), rendering generally good performance (with R2 = 0.902 against 601 experimental data points). The developed ML-based IFT equation can be readily applied and implemented in reservoir simulations and other UHS applications.