← AI security in plain English

LLM08

OWASP Top 10 for LLMs / Vector and Embedding Weaknesses

Flaws in how a system generates, stores, or retrieves the numeric representations behind retrieval-augmented generation let attackers poison the knowledge store or pull out information they should not access.

Think of it likeA home filing cabinet with a broken lock, where anyone can slip in fake documents or read the family's private folders.

In plain English

AI systems that look up stored knowledge can be attacked through that memory bank, letting bad actors plant false facts or steal what is filed there.