Artificial Intelligence Is Not an Intern: Authoritative Prompt Parenting for the Supervision of Generative AI by Engineers, Scientists, and Health Professionals

Abstract

Generative artificial intelligence does not hold professional responsibility, and it may not reliably preserve user instructions over extended interactions. In response, we present “authoritative prompt parenting,” as a supervisory heuristic drawn from developmental science. The approach comprises four standing-instruction controls, including: tiered rules; a visible tripwire; a counter-indexed heartbeat; and a traverse/paraphrase read-receipt that tests whether designated instruction fragments are accessible and whether core rules can be restated. In a point-in-time paired demonstration using four Large Language Model (LLM) systems, the specified initial receipt was retrieved from the installed instruction file, and paired controls did not. While the approach does not confirm whole-file attention, human-like comprehension, or continued adherence, the approach does prompt human verification – and a conservative restart when a check fails – to support human professional accountability.