In this talk I show that theoretical predictions of MSSM scalar masses have uncertainties from interactions of the hidden sector which have been overlooked in all prior work. These interactions can contribute such large corrections to the renormalization of superpartner masses that predictions from commonly used spectrum codes (such as ISAJET, SPHENO, SOFTSUSY, SUSPECT) may be completely wrong. The size of the new effects depends on the strength of interactions in the hidden sector, and can easily be of order 1. As an example, in MSUGRA with unified masses (m0, m1/2) at the GUT scale squarks may end up lighter than sleptons at the TeV scale.