Title: The AGN-galaxy-halo connection: The distribution of AGN host halo masses to z=2.5

Abstract: It is widely reported, based on clustering measurements of observed active galactic nuclei (AGN) samples, that AGN reside in similar mass host dark matter halos across the bulk of cosmic time, with log $M/M_\odot$~12.5-13.0 to z~2.5. We show that this is due in part to the AGN fraction in galaxies rising with increasing stellar mass, combined with AGN observational selection effects that exacerbate this trend. Here, we use AGN specific accretion rate distribution functions determined as a function of stellar mass and redshift for star-forming and quiescent galaxies separately, combined with the latest galaxy-halo connection models, to determine the parent and sub-halo mass distribution function of AGN to various observational limits. We find that while the median (sub-)halo mass of AGN, $\approx10^{12}M_\odot$, is fairly constant with luminosity, specific accretion rate, and redshift, the full halo mass distribution function is broad, spanning several orders of magnitude. We show that widely used methods to infer a typical dark matter halo mass based on an observed AGN clustering amplitude can result in biased, systematically high host halo masses. While the AGN satellite fraction rises with increasing parent halo mass, we find that the central galaxy is often not an AGN. Our results elucidate the physical causes for the apparent uniformity of AGN host halos across cosmic time and underscore the importance of accounting for AGN selection biases when interpreting observational AGN clustering results. We further show that AGN clustering is most easily interpreted in terms of the relative bias to galaxy samples, not from absolute bias measurements alone.