Prompt Injection Doesn't Come from Your Users

«Your team added content filtering. You're scanning user messages for injection patterns before they reach the model. You feel reasonably secure about the input path. Meanwhile, the database record your agent queried thi...»