Using the results of recent optical surveys, we conclude that the nondetection of quasars down to faint magnitudes implies a significant flattening of the high-redshift (z $ 6) optical active galactic nucleus (AGN ) luminosity function for M 1450 k À24:7. We find that all the data are consistent with a faint-end slope for the optical AGN luminosity function of ¼ À2:2 and À2.8, at the 90% and 99% confidence level, respectively, which is flatter than the bright-end slope of 0 $ À3:2. We also show that X-ray deep surveys have probed even fainter magnitudes than the optical ones, yielding more significant constraints on the shallow faint-end slope of the optical luminosity function. The inclusion of type II AGN candidates, detected in the Chandra Deep Fields, hints toward a higher normalization for the total AGN luminosity function, if these sources lie at 5 P z P 6:5. We then discuss simple theoretical models of AGN formation and evolution in the context of cold dark matter cosmology. The comparison with the total AGN luminosity function favors a redshift-dependent relation between black hole and dark matter halo masses of the type M / M halo , with 1:3 P P 1:7, compatible with independent studies from statistical analysis and rotation curve measurements. Finally, we compute the quasar contribution to reionization to be P9% at z $ 6, up to $30% when integrated within 5:5 P z P 6:5, significantly smaller than that from galaxies.