GhostApproval: A trust-boundary gap in AI coding assistants

Uncovering a category-level blind spot in modern AI coding assistants, and why the Human-in-the-Loop safety model fails against this classic threat