The short GRB 120323A had the highest flux ever detected with the Fermi/GBM. Here we study its remarkable spectral properties and their evolution using two spectral models: (i) a single emission component scenario, where the spectrum is modeled by the empirical Band function, and (ii) a two component scenario, where thermal (Planck-like) emission is observed simultaneously with a non-thermal component (a Band function). We find that the latter model fits the integrated burst spectrum significantly better than the former, and that their respective spectral parameters are dramatically different: when fit with a Band function only, the Epeak of the event is unusually soft for a short GRB, while adding a thermal component leads to more typical short GRB values. Our time-resolved spectral analysis produces similar results. We argue here that the two-component model is the preferred interpretation for GRB 120323A, based on: (i) the values and evolution of the Band function parameters of the two component scenario, which are more typical for a short GRB, and (ii) the appearance in the data of a significant hardness-intensity correlation, commonly found in GRBs, when we employee two-component model fits; the correlation is non-existent in the Band-only fits. GRB 110721A, a long burst with an intense photospheric emission, exhibits the exact same behavior. We conclude that GRB 120323A has a strong photospheric emission contribution, first time observed in a short GRB. Magnetic dissipation models are difficult to reconcile with these results, which instead favor photospheric thermal emission and fast cooling synchrotron radiation from internal shocks. Finally, we derive a possibly universal hardness-luminosity relation in the source frame using a larger set of GRBs L,i=(1.59+/-0.84).10^50 (Epeak,i)^(1.33+/-0.07) erg/s), which could be used as a possible redshift estimator for cosmology.