Should scientific expertise take into account the interests or values of society? According to an argument from inductive risk (Rudner, Douglas), non-epistemic values have to enter the picture. This threatens the so-called linear model of expertise. We argue that this model should be amended: scientific expertise can be value-free in the sense that scientists will not choose the non-epistemic values (but not in the sense that they will not use non-epistemic values in their judgments), on the condition that these non-epistemic values are provided by the requiring institution, and thus externalized. We consider the case of the IPCC.