When AI Advises Your Boss, the Warnings Can Quietly Disappear

AI is now woven into how big decisions get made at work. It drafts the reports, suggests the options, and scores the pros and cons that executives read before saying yes. A new 25-page paper by Richard Hill, published on arXiv, argues that something is slipping through the cracks. Each human and each AI step looks perfectly competent on its own, yet the final decision can rest on weaker ground than anyone realises.

The paper names the culprit: 'qualification attrition.' Think of a contract photocopied again and again, or a message passed along a game of telephone. Every handoff — a summary, a translation, an AI rewrite, a tidy dashboard — drops a caveat, a doubt, or an assumption that mattered. By the time the decision reaches the person who signs off, the reasoning behind it may no longer be visible, understandable, or challengeable. Hill calls this the 'formation–authorisation gap': the gap between what you'd need to justify a decision and what the approver can actually still see.

His fix isn't to ban AI. He proposes four practical habits for organisations. First, boundary setting — deciding clearly what AI is and isn't allowed to touch. Second, interpretive challenge — keeping a human who can question how the AI read the situation. Third, reliance calibration — knowing when to trust the output and when to double-check. Fourth, authorisation with answerability — the person signing off genuinely owns the call. He then links these to risks of bad judgement and unclear accountability.

Here's the honest catch: this is a theory paper, not a study. There are no experiments, no company case studies, no numbers proving how often this happens. It's a framework — a smart warning that needs testing in real workplaces. Still, it names something many employees already sense: when AI smooths the path to a decision, the bumps that made you think carefully can get paved over. If your boss's next big call was shaped by an AI summary, someone should still be able to explain the reasoning out loud — and be challenged on it.