Two gaps in every hosted AI call: which model ran, and who saw the prompt. What I run to close both, and where each one still falls short.

A developer outlines two gaps in every hosted AI API call — verifying which model actually ran and who saw the prompt — and describes the tooling they run to close both, along with where each approach still falls short. The writeup covers open-source and privacy-focused methods for auditing LLM providers rather than trusting them on faith.