# AI Usage Policy: Rules People Follow, Not a Ban

An AI usage policy people follow: three data levels, a named list of approved tools, exceptions answered in two days, and a one-page version for employees.

Source: https://node8.ai/ai-usage-policy/

---

Governance & control

# An AI policy that bans everything gets bypassed on day one

Most AI policies are written to protect the company on paper, so they forbid almost everything. People keep using AI anyway, on personal accounts, and stop telling anyone. A policy that works starts from what is allowed.

Three data levelsA named list of approved toolsExceptions answered in two daysClear human review rulesA one-page version for employees

Book a 20-minute call [Ask on WhatsApp](https://wa.me/16465352924?text=Hi%20Boaz%2C%20I%20saw%20the%20AI%20usage%20policy%20page%20on%20your%20site%20and%20I'd%20love%20some%20details.)

Why policies get ignored

## Four patterns behind an AI policy nobody follows

### It bans instead of guiding

When everything is forbidden, each employee decides alone what is fine. Usage moves to personal accounts, and you end up with less control than before.

### It is written for lawyers

Twelve pages of defined terms, and the person at the keyboard still can’t tell whether this contract may go into this tool. A policy nobody reads is shadow AI with a PDF.

### There is no list of approved tools

No single place says which tools are allowed, for which data, signed in how. So people guess, and the guesses differ by team.

### There is no way to ask for an exception

When someone needs a tool that isn’t covered, there is nobody to ask, or the answer takes a month. So they go around the policy. An exception answered in two days prevents ten bypasses.

What we do

## Legal signs off. We write what happens on Monday morning.

We are not lawyers and don’t give legal advice. We bring the operational side: which tool, which data, who approves, and how fast.

### 1\. Map what is actually in use

First we find out which AI tools people already use, and with which data. Rules written for the company you wish you had get ignored.

-   **Input** — Sign-in logs from Microsoft Entra ID or Google Workspace, AI subscriptions in expenses, and a five-minute staff survey
-   **The AI** — Groups the answers by tool, team and type of data involved
-   **Output** — A one-page usage map, including the tools nobody approved
-   **Checkpoint** — Security and team leads confirm it reflects reality, with no blame attached

### 2\. Draft the data levels and the tools list

Three levels of data, described in plain words with examples from your own documents, and a named list saying which approved tool may touch which level.

-   **Input** — Your existing data classification, vendor contracts and the usage map
-   **The AI** — Drafts the three levels with real examples and matches each approved tool to a level
-   **Output** — A draft classification and tools list
-   **Checkpoint** — Security, legal and HR review it together before anything is published

### 3\. Build the exception path and review rules

Who approves exceptions, and how fast. Plus a short list of what a person must check before it leaves the company.

-   **Input** — An exception request through a short form in Slack, Teams or Jira
-   **The AI** — Checks the request against the policy and tools list and drafts a recommendation
-   **Output** — A decision draft with the open questions already answered
-   **Checkpoint** — A named approver decides within two business days

### 4\. Roll it out and keep it current

A one-page version taught in a workshop, and a monthly look at what the exceptions say about the rules.

-   **Input** — The exception log and new tool requests from the month
-   **The AI** — Summarizes patterns, such as the same exception requested by five people
-   **Output** — A short list of proposed changes to the policy or tools list
-   **Checkpoint** — The policy owner decides what changes, and the tools list is updated

The outline

## What the policy actually contains

Ten short sections, each with an example line written the way an employee would read it.

Section

What it says

Example line

Purpose and scope

Who and what the policy covers

“This applies to everyone doing company work with AI, on any device.”

What is allowed

The default is yes, within the rules

“Use approved AI tools to draft, summarize, research and analyze Level 1 and Level 2 data.”

Data levels

Three levels with examples

“Check the level before you paste. If you aren’t sure, treat it as Level 3.”

Approved tools

The named list, its owner and its review date

“Approved for Level 2 is our company ChatGPT Business workspace, signed in with your work account.”

Human review

What a person checks before it leaves

“Anything sent to a customer, regulator or the public is reviewed and owned by a person.”

Agents and automation

Agents that act on systems need an owner

“Every agent that writes to a company system has a named owner and a registry entry.”

Exceptions

Who approves, how to ask, how fast

“Ask in the #ai-help channel. You get an answer within two business days.”

Transparency

When to say AI was involved

“Customers are told when they are talking to an AI agent rather than a person.”

Incidents

What to do when data went to the wrong place

“Tell security the same day. Reporting a mistake is never what gets you in trouble.”

Ownership and review

Who owns the policy and how often it changes

“The policy owner reviews the tools list every quarter and after any incident.”

Data classification

## Three levels, not seven

A scheme only works if someone can apply it in the five seconds before they paste.

Level

Examples

Which AI tools may touch it

Level 1, Public

Website copy, published reports, press releases, job ads, public filings

Any approved tool

Level 2, Internal

Internal documents, meeting notes, project plans, internal metrics, routine customer emails without sensitive detail

Approved company workspaces only, such as ChatGPT Business or Enterprise, Claude Team or Enterprise, Microsoft 365 Copilot or Gemini Enterprise, signed in with a work account

Level 3, Restricted

Customer and employee personal data, health information, payment data, contracts under NDA, financials before release, source code and credentials

Only tools and agents security has approved for this level, with access controls and logging. Credentials and card numbers never go into AI

Regulation and frameworks

## Where the rules come from

The policy maps to the frameworks your customers and auditors already use. This is an overview, not legal advice.

### GDPR

Personal data in a prompt is still personal data. You need a lawful basis, a data processing agreement with the vendor and a view of where the data is processed.

### EU AI Act

If you operate in or serve the EU, Article 4 covers AI literacy: companies using AI are expected to take measures on their staff’s AI knowledge. Transparency obligations under Article 50 have applied since August 2026, for example telling people when they are dealing with an AI system. A policy plus training is the practical way to show both.

### HIPAA, where health data is in play

Protected health information only goes into tools covered by a business associate agreement and configured for it.

### SOC 2 and ISO/IEC 42001

When a customer’s security questionnaire asks about AI, the policy, the tools list and the exception log are what you send. ISO/IEC 42001, the AI management system standard, and the NIST AI RMF both build on documented AI policies, and this one is a usable first block.

The offer

## A policy in a couple of weeks, drafted with legal, security and HR

Two working sessions and a review round, with the people who will have to stand behind it. You end with four things:

### The full policy

A short operational draft, built with your legal, security and HR leads, ready for your counsel to approve.

### A one-page employee version

What’s allowed, the three levels, the tools list and how to ask for an exception. The page people actually read.

### The tools list and exception path

Named tools per data level, a named approver, a request form and a two-day answer target.

### Rollout through the workshop

The policy is taught on people’s own work in an AI workshop, and AI champions in each team become the first place people ask.

In our [AI training and engineering enablement](https://node8.ai/case-studies/ai-training-engineering-enablement/) engagement, an AI Acceptable Use Policy and responsible-use guardrails were part of the program from the start, alongside company-wide training, and gave the company governance mature enough to satisfy both legal and ownership.

When this comes up

## If two of these sound familiar, it’s worth a call

-   Security blocked AI tools and people moved to personal accounts
-   A customer or auditor asked for your AI policy
-   You are about to roll out ChatGPT, Claude, Copilot or Gemini company-wide
-   There is an old policy nobody has read

FAQ

## AI usage policy questions, answered

Do you write the policy for us?

We write the operational draft and work through it with your legal, security and HR leads. Your counsel approves the final text. We are not lawyers, we don’t give legal advice, and we don’t replace yours.

Do you understand our field?

No, you do. You know which data is sensitive in your business and which mistakes would hurt. We bring the structure, the examples from other companies and the practical view of what happens when a policy meets a Monday morning.

What data should never go into AI?

Credentials, passwords and card numbers, in any tool. Beyond that it depends on the tool: customer personal data, health information and confidential contracts belong in Level 3, which only tools approved for it by your security team may touch.

Should we just block ChatGPT and the others?

Blocking without an approved alternative moves usage to personal phones and accounts, where you can see nothing. It is usually safer to approve a business workspace with company sign-in, set the data rules, and block the consumer versions for company data.

Does the EU AI Act apply to us?

If you operate in the EU or your AI use reaches people there, parts of it may. For most companies using AI tools rather than building them, the relevant parts are AI literacy and transparency. We point out where it is likely to apply; your counsel confirms.

How is a usage policy different from AI governance?

The usage policy sets the rules for people: what to paste where, and what needs review. AI governance covers the agents themselves: the registry, the owner of each agent, its permissions and its audit log. You need both, and they should point to each other.

Where this leads

## The pages that continue from here

[AI governance](https://node8.ai/ai-governance)[Security](https://node8.ai/security)[AI implementation](https://node8.ai/ai-implementation)[AI workshops](https://node8.ai/learn)

## What would your people need to know on Monday morning?

A short call, no deck. We will look at what you have today, what people are actually using, and say honestly whether we can add anything to it.

Book a 20-minute call [Ask on WhatsApp](https://wa.me/16465352924?text=Hi%20Boaz%2C%20I%20saw%20the%20AI%20usage%20policy%20page%20on%20your%20site%20and%20I'd%20love%20some%20details.)
