Definition
An attack targeting the searchable knowledge database (called a vector store or vector database) that powers AI applications. By inserting specially crafted content into that database, an attacker can manipulate the information the AI retrieves and therefore control or corrupt the AI's answers. It is analogous to tampering with the reference library an AI consults before speaking.
Why it matters
Vector stores are the authoritative knowledge backbone of most enterprise AI assistants. Compromising one can silently poison every query the AI handles, affecting business decisions, compliance documents, or customer interactions at scale.