Agent Memory Attacks: Poison a Shared Store, Read Other Users' Sessions
Agent memory attacks, hands-on: poison a shared knowledge base so every user gets your answer, and read other users' saved notes by guessing sequential session IDs.
Agent memory attacks, hands-on: poison a shared knowledge base so every user gets your answer, and read other users' saved notes by guessing sequential session IDs.
Direct prompt injection, step by step: pull an AI agent's system prompt and database creds past an output filter, then hijack a second agent's goal. Lab only.
Indirect prompt injection by cross-document fragmentation: split one instruction across two clean files, pass the upload scanner, and leak its MinIO keys.