Jennifer Radden has drawn attention to two features of delusion, ambivalence and subjectivity, which are problematic for theories of delusion that treat delusions as empirical beliefs. She argues for an ‘attitude’ theory of delusion. I argue that once the cognitive architecture of delusion formation is properly described the debate between doxastic and attitude theorists loses its edge. That architecture suggests that delusions are produced by activity in the ‘default mode network’ unsupervised by networks required for decontextualized processing. The cognitive properties of these networks explain the features of delusion which generate the debate between doxastic and attitude theorists.