Recently, Brodsky, Hwang and Schmidt have proposed a new mechanism that gives a transverse spin symmetry at leading twist in semiinclusive deep-inelastic scattering. I show that the new mechanism is compatible with factorization and is due to an transverse-spin asymmetry in the k T distribution of quarks in a hadron (the "Sivers asymmetry"). An earlier proof that the Sivers asymmetry vanishes because of time-reversal invariance is invalidated by the path-ordered exponential of the gluon field in the operator definition of parton densities. Instead, the time-reversal argument shows that the Sivers asymmetry is reversed in sign in hadron-induced hard processes (e.g., Drell-Yan), thereby violating naive universality of parton densities. Previous phenomenology with time-reversal-odd parton densities is therefore validated.