The Unread Instruction

Abstract

Every enterprise AI deployment runs on an instruction written by the organisation that uses it. This instruction controls much of what the AI will and will not produce, and whoever has access can change it at any time. This paper asked whether the major frameworks that tell boards how to oversee AI require anyone to know about this instruction. We read nine public frameworks in their current versions on 4 July 2026. Across all twenty-seven answers, not one was yes. None of the frameworks require anyone to know who wrote the system prompt, when it was last changed, or how it affects answers. The phrase "system prompt" did not appear in any document.