We infer the radial gravitational acceleration around isolated galaxy-galaxy lenses from KiDS-1000 weak lensing convergence measurements. We extend the radial acceleration relation (RAR), conventionally measured using galaxy rotation curves, by 2 decades in acceleration. We compare the measurements with mock lensing "observations" of the MICE N-body + semi-analytic galaxy formation simulation, and the BAHAMAS cosmological hydrodynamical simulations. We find MICE predictions to be in excellent agreement with our measurements, but these simulations fail to reproduce y expected bias in the stellar-to-halo-mass ratio of isolated galaxies. Conversely, the BAHAMAS simulations do reproduce this bias, but this causes the RAR to shift well away from the measured trend. We also compare to predictions of two modified gravity models, MOND and Emergent Gravity, finding these to be consistent with the overall measured RAR. However, we find a significant difference in the RAR of early- and late-type galaxies, which these theories do not predict, but is straightforwardly accommodated in dark matter models. Our interpretation of our measurements is limited primarily by uncertainty in the distribution of hot, circumgalactic gas.