Doing what everyone should be doing and using the LLM for what it’s good at (natural language processing) and not for what it’s bad at (decision making). In a high stakes context like this, trusting an LLM for anything more is negligent — arguably criminally so.