In our recent experimental tests of Bell’s inequality, for the first time, we used observations of astronomical sources to randomly choose measurement settings for polarization-entangled photons sent through free space to two distant detectors. In our 2017 pilot test in Vienna, Austria, we used the color of light from Milky Way stars, and in 2018, we performed the first Cosmic Bell test in the Canary Islands using light from high redshift quasars. In both sets of tests, we observed statistically significant violation of Bell’s inequality, implying that John Bell’s very reasonable assumptions about the world cannot all be true in nature. These assumptions include realism/determinism, locality, and experimental freedom of choice. Our tests aim to put tension on the latter assumption, arguably the most subtle of Bell’s axioms, which holds that each detector’s measurement choices are completely free of any non-quantum degrees of freedom in the causal past of the experiment that could also affect the measurement outcomes. Since the nearest star in the pilot test was ~600 light years away, given our assumptions, the observed Bell violation implies that any non-quantum explanations for entanglement must have been in place prior to ~600 years ago, an improvement of ~16 orders of magnitude compared to previous tests. Similarly, since the nearest quasar in our best experimental run was 7.8 billion light years away, any such mechanism is relegated an additional ~6 orders of magnitude into the cosmically distant past, corresponding to excluding non-quantum explanations from 96% of the space-time volume in the causal past of the experiment. In addition to exploring how free our experimental choices are while investigating the fundamental nature of reality in the subatomic world, such foundational tests are relevant to whether practical quantum encryption schemes will ultimately be as secure as many researchers believe. Our other new theoretical work shows that relaxing freedom of choice by a relatively small amount can still reproduce the quantum predictions, making the associated loophole easier to exploit than other well known entanglement test loopholes.