International Association
for Cryptologic Research

In the algebraic group model (AGM), an adversary has to return with each group element a linear representation with respect to input group elements. In many groups, it is easy to sample group elements obliviously without knowing such linear representations. Since the AGM does not model this, it can be used to prove the security of spurious knowledge assumptions. We show several well-known zk-SNARKs use such assumptions. We propose AGM with oblivious sampling (AGMOS), an AGM variant where the adversary can access an oracle that allows sampling group elements obliviously from some distribution. We show that AGM and AGMOS are different by studying the family of ``total knowledge-of-exponent'' assumptions, showing that they are all secure in the AGM, but most are not secure in the AGMOS if the DL holds. We show an important separation in the case of the KZG commitment scheme. We show that many known AGM reductions go through also in the AGMOS, assuming a novel falsifiable assumption TOFR. We prove that TOFR is secure in a version of GGM with oblivious sampling.