We investigate the direct detection phenomenology of a class of dark matter (DM) models in which DM does not directly interact with nuclei, but rather the products of its annihilation do. When these annihilation products are very light compared to the DM mass, the scattering in direct detection experiments is controlled by relativistic kinematics. This results in a distinctive recoil spectrum, a non-standard and or even absent annual modulation, and the ability to probe DM masses as low as a ∼10 MeV. We use current LUX data to show that experimental sensitivity to thermal relic annihilation cross sections has already been reached in a class of models. Moreover, the compatibility of dark matter direct detection experiments can be compared directly in Emin space without making assumptions about DM astrophysics, mass, or scattering form factors. Lastly, when DM has direct couplings to nuclei, the limit from annihilation to relativistic particles in the Sun can be stronger than that of conventional non-relativistic direct detection by more than three orders of magnitude for masses in a 2-7 GeV window.Introduction -While very little is known about Dark Matter (DM), its cosmological abundance is experimentally quite well-determined: Ω CDM h 2 = 0.1199 ± 0.0027 [1]. An appealing framework for understanding the relic abundance of Dark Matter (DM) is thermal freeze-out [2]. Number-changing interactions in the early universe, XX ↔ (SM)SM keep DM in thermal equilibrium with the SM bath, until the rate of these annihilation processes drops below the rate of Hubble expansion. After this point the abundance of DM is essentially fixed at, Ω CDM h 2 0.12 6 × 10 −26 cm 3 s −1 / σ ann v rel , singling out a characteristic annihilation cross section σ ann v rel for thermally produced DM to yield the observed abundance. This scenario is attractive in that it provides a simple and elegant framework for the relic abundance that can be tested in a variety of ways, including direct detection (DD) [3]. However, current constraints from DD rule out many of the simplest models of thermal relic DM, which may indicate a modification of the above picture.In this paper we investigate a modification of thermal DM which alleviates the tension between DD constraints and the thermal relic hypothesis, while making unique predictions for DD. In particular, we take the abundance of DM, X, to be determined by the annihilation process XX ↔ Y Y , where Y is a much lighter dark sector species. The interactions of the dark sector state Y with ordinary nuclei allows for a unique test of the scenario at DD experiments. The resulting DD phenomenology of this class of models is distinctive, owing to the fact that (1) the scattering partner of the nucleus is relativistic, rendering the kinematics of scattering completely different and (2) it is the flux of the scattering partner Y that determines the rate of events at a detector rather than X. Both of these features have novel consequences not considered in the literature of "model-independent" direct detecti...