It is widely accepted that observations at mid-infrared (mid-IR) wavelengths enable the selection of galaxies with nuclear activity, which may not be revealed even in the deepest X-ray surveys. Many mid-IR color-color criteria have been explored to accomplish this goal and tested thoroughly in the literature. Besides missing many low-luminosity active galactic nuclei (AGN), one of the main conclusions is that, with increasing redshift, the contamination by non-active galaxies becomes significant (especially at z 2.5). This is problematic for the study of the AGN phenomenon in the early Universe, the main goal of many of the current and future deep extra-galactic surveys. In this work new near-and mid-IR color diagnostics are explored, aiming for improved efficiency -better completeness and less contamination -in selecting AGN out to very high redshifts. We restrict our study to the James Webb Space Telescope wavelength range (0.6-27 µm). The criteria are created based on the predictions by state-of-the-art galaxy and AGN templates covering a wide variety of galaxy properties, and tested against control samples with deep multi-wavelength coverage (ranging from the X-rays to radio frequencies). We show that the colors K s − [4.5], [4.5] − [8.0], and [8.0] − [24]are ideal as AGN/non-AGN diagnostics at, respectively, z 1, 1 z 2.5, and z 2.5 − 3. However, when the source redshift is unknown, these colors should be combined. We thus develop an improved IR criterion (using K and IRAC bands, KI) as a new alternative at z 2.5. KI does not show improved completeness (50-60% overall) in comparison to commonly used IRAC-based AGN criteria, but is less affected by non-AGN contamination (revealing a >50-90% level of successful AGN selection). We also propose KIM (using K, IRAC, and MIPS-24 µm bands, KIM), which aims to select AGN hosts from local distances to as far back as the end of reionization (0 < z 7) with reduced non-AGN contamination. However, the necessary testing-constraints and the small control-sample sizes prevent the confirmation of its improved efficiency at z 2.5. Overall, KIM shows a ∼30-40% completeness and a >70-90% level of successful AGN selection. KI and KIM are built to be reliable against a ∼10-20% error in flux, are based on existing filters, and are suitable for immediate use.