This paper formalizes the concept of entity-level deepfakes—synthetic constructs that maintain persistent digital identities across platforms, participate in fabricated reference networks, and are designed to be indistinguishable from authentic entities. Unlike media deepfakes that target perception, entity-level deepfakes attack innovation infrastructure: patent databases, academic citation systems, and AI training corpora. We introduce formal definitions for Synthetic Saturation, Recursive Corpus Corruption, and Epistemic Infrastructure, and propose three implementable countermeasures: cryptographic invention timestamping, a Synthetic Density Index (SDI), and a Verified Training Corpus Standard (VTCS). The analysis is grounded in operational data from Helix Fabric, a deployed synthetic organization detection system scanning 1,700+ targets with 15 signal types achieving composite detection confidence exceeding 0.85. This is paper #6 in a constitutional AI governance research program.