Recent analyses show that $\Lambda$CDM-based models optimised to reproduce the clustering of massive galaxies overestimate their gravitational lensing by about 30\%, the so-called lensing is low problem. Using a state-of-the-art hydrodynamical simulation, we show that this discrepancy reflects shortcomings in standard galaxy-halo connection models rather than tensions within the $\Lambda$CDM paradigm itself. Specifically, this problem results from ignoring a variety of galaxy formation effects, including assembly bias, segregation of satellite galaxies relative to dark matter, and baryonic effects on the matter distribution. All these effects contribute towards overestimating gravitational lensing and, when combined, explain the amplitude and scale dependence of the lensing is low problem. We conclude that simplistic galaxy-halo connection models are inadequate to interpret clustering and lensing simultaneously, and that it is crucial to employ more sophisticated models for the upcoming generation of large-scale surveys.