Giving a local LLM full VM access showed me why we need better AI guardrails

The prompt injection is coming from inside the house