Survey after survey finds the same thing: many employees use generative AI tools at work, and many of them use personal accounts their employer doesn't know about. This is shadow AI. The risk is real: confidential documents, customer data and source code pasted into consumer tools with unclear data handling. But blanket bans tend to push usage further underground. A better approach combines a short policy, approved tools and light controls.

Why bans don't work

The tools are useful, free or cheap, and available on any phone. If the approved route is "no", people will use their own devices. You lose visibility and gain no protection. The goal is to make the safe way the easy way.

What the policy should cover

Keep it to one or two pages. People don't read long policies.

1. Purpose and scope

Which tools it covers (chat assistants, coding assistants, AI features inside existing software, image and audio generators) and who it applies to (employees, contractors).

2. Approved tools

A short list of tools approved for work use, with enterprise agreements that cover data handling: no training on your data, retention limits, security commitments. Explain how to request a new tool.

3. What data may go into which tool

The most important section. Tie it to your existing data classification:

Data class Approved enterprise AI tools Consumer or unapproved tools
Public Yes Yes
Internal Yes No
Confidential (customer data, financials, source code) Yes, if the tool is approved for that class No
Restricted (national ID numbers, health data, credentials, unreleased results) Only specifically approved systems Never

4. Responsible use

  • You are responsible for the output. Check facts, code and citations before using them.
  • Disclose AI use where it matters: customer-facing content, regulated advice, academic or legal work.
  • No automated decisions about people (hiring, performance, credit) without an approved process and human review.
  • Respect intellectual property and third-party terms.
  • Don't use AI to impersonate people or create misleading content.

5. Building with AI

Anyone building AI into products or processes, including agents and automations, follows the AI intake and risk assessment process.

6. Reporting

How to report a mistake, such as pasting confidential data into the wrong tool. Make reporting blame-free so people actually do it.

Back the policy with controls

  • Provide good approved tools. This is the single biggest lever. If the enterprise tool is as good as the consumer one, most people will switch.
  • Single sign-on to approved AI tools, so access ends when someone leaves.
  • Discovery. Use your secure web gateway, CASB or DNS logs to see which AI services are being used. Use the data to decide what to approve, not just what to block.
  • Data loss prevention rules for the most sensitive data types going to AI sites.
  • Browser and endpoint controls for high-risk teams where needed.
  • Review AI features in existing SaaS. Many tools you already use have added AI features. Check their data terms.

Train in minutes, not hours

A 10-minute module with three or four realistic examples works better than an hour of theory: "Can I paste this customer email into the assistant to draft a reply?" Show the right answer and the approved tool for doing it.

Measure

Track adoption of approved tools, discovered unapproved usage over time, DLP events and policy exceptions. Falling shadow usage with rising approved usage means the approach is working.

Learn more

AI governance, acceptable use and data protection for AI are covered in our free AI Security & Governance course. For privacy obligations when AI processes personal data, see our privacy course.

Written and reviewed by the DigiTrust Asia editorial team. How we research, review and correct articles.

General information, not legal advice. Laws and official guidance change; check the current text before acting.

Build AI governance skills

AI Security & Governance is a free course with a final exam and certificate.

Create a free account