What happens when everyone has their own AI assistant?

A hypothetical example: Management has yet to decide how the company should use AI. An employee nevertheless uses an AI assistant to shorten a long customer brief before a meeting. The task becomes easier, but nobody has established whether the brief may be shared with the tool.

This kind of use without the company’s oversight or approval is often called shadow AI. It is worth examining both the information being shared and the work the employee is trying to get done.

Start with the task

A conversation about what takes unnecessarily long gives you a concrete starting point. Perhaps it is meeting notes. Perhaps it is searching internal documents for answers.

Ask employees to describe the task and demonstrate the process using material they may share. The aim is to understand the need and identify where they are unsure. A list of tools becomes more useful when you know what people use them for.

When several people tackle the same task in different ways, that is an obvious place for a shared solution. It still needs to work in practice. An approved tool that does not do the job sends people straight back to whatever they were using before.

Make the rules specific

Employees need to be able to tell whether approval covers their task. Permission to draft copy from public information says nothing about using confidential customer documents.

A short guide can describe approved tasks and the limits that come with them. It should answer questions such as:

• Which tool and work account should be used for the task?
• What information may be shared, and what must be kept out?
• What should an employee check before using the result?
• Who can help when the task falls outside the guide?

A constructed example of an answer that sounds correct but contains fabricated information can help. It makes the need for checking easier to understand than a general instruction to be careful.

Who handles requests and mistakes?

Someone needs to say yes or no to new tools, and everyone needs to know who. Employees explain the need and learn what happens next.

If an employee shares information they should not have shared, they need to know who to tell. The person responsible investigates what was shared and what the company can do to limit the consequences.

If you introduce an internal AI assistant, test which documents it can see. Calling it internal does not control who gets access.

Follow up on whether the approved ways of working actually help. If employees still copy information between several systems to complete the same task, there is a concrete problem left to solve.

Shared rules suit companies where employees already use AI or want to get started. They give employees a clear starting point and can save the company time spent answering recurring questions and dealing with mistakes.