---
id: "framework-accountability-rules"
type: "framework"
source_timestamps: ["§ 2. Make accountability explicit and personal for employees and agents."]
tags: ["governance", "accountability", "risk-management"]
related: ["framework-responsible-human-ai-collaboration", "action-define-decision-rights"]
steps: ["Decision rights: Define exactly what the agent may do autonomously versus what requires explicit human approval.", "\\\"Escalation: Define what triggers a review", "who intervenes", "and who bears the cost of delay or error.\\\"", "\\\"Consequences: Define what happens when the agent fails", "how accountability is carried forward", "and who is responsible for monitoring and improving agent performance.\\\""]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-ext-16-dont-treat-agents-like-employees"
sourceUrl: "https://hbr.org/2026/05/research-why-you-shouldnt-treat-ai-agents-like-employees"
sourceTitle: "Research: Why You Shouldn’t Treat AI Agents Like Employees"
---
# Rules of Collaboration for Humans and Agents

**A sub-framework of [[framework-responsible-human-ai-collaboration]] (Step 2)** for establishing explicit, personal accountability when deploying AI agents — especially critical in highly regulated environments. Because agentic AI can handle far more flexible situations than traditional automation, organizations must establish clear boundaries along **three fronts**:

1. **Decision rights** — Define exactly what the agent is allowed to do **autonomously** versus what requires **explicit human approval**.
2. **Escalation** — Define what **triggers a review**, **who intervenes**, and **who bears the cost** of delay or error.
3. **Consequences** — Define what happens when the agent fails, how **accountability is carried forward**, and who is responsible for **monitoring and improving** agent performance.

This ensures that when AI contributes to an outcome, the responsible human is acutely aware of their accountability for that output — directly dismantling [[concept-accountability-blurring]] and the measured shift in [[claim-accountability-shift-d6]]. It is put into practice by [[action-define-decision-rights]]. An unresolved legal dimension is captured in [[question-legal-accountability]].