How I Detect Multi-Turn Prompt Injections Without ML

Every LLM firewall I've seen analyzes each message in isolation. Send a prompt, get a score, block or...