This paper presents secure network provenance (SNP), a novel technique for tracking down compromised nodes in a distributed system and assessing the damage that they may have caused to other nodes. SNP enables operators to ask the system why it is in a certain state – for example, why a suspicious routing table entry is present on a certain router, or where a given cache entry originated. SNP is robust to manipulation; its tamper-evident properties ensure that operators can detect when compromised nodes lie or falsely implicate correct nodes. Thus, compromised nodes can at worst refuse to participate, making their presence evident to operators. We describe an algorithm for answering SNP queries, as well as a proof-of-concept implementation.