Isn’t prompt injection a massive problem here? User content could ask the LLM to ignore it. Or potentially even ask for the “counter speech” to include any pri

Isn’t prompt injection a massive problem here? User content could ask the LLM to ignore it. Or potentially even ask for the “counter speech” to include any private information the LLM has access to.