This work addresses the problem of learning from large collections of data with privacy guarantees. The compressive learning framework proposes to deal with the large scale of datasets by compressing them into a single vector of generalized random moments, from which the learning task is then performed. We show that a simple perturbation of this mechanism with additive noise is sufficient to satisfy differential privacy, a well established formalism for defining and quantifying the privacy of a random mechanism. We combine this with a feature subsampling mechanism, which reduces the computational cost without damaging privacy. The framework is applied to the tasks of Gaussian modeling, k-means clustering and principal component analysis (PCA), for which sharp privacy bounds are derived. Empirically, the quality (for subsequent learning) of the compressed representation produced by our mechanism is strongly related with the induced noise level, for which we give analytical expressions.