Understanding data permissions is key with Copilot

Microsofts various AI driven offerings (or “Copilots”) have definitely gained a lot of interest over the past few months. Central to the Copilot brand, is Copilot for Microsoft 365. Copilot for Microsoft 365 is one of the easiest solutions in Microsofts overall Copilot initiative to understand from a value and use-case perspective. While there are undoubtedly countless business scenarios where specific models can drive true, meaningful change across almost every industry, the more advananced, specific uses for Large Language Models (LLMs) generally have a longer time to value than something like Copilot for Microsoft 365.

Copilot for Microsoft 365 brings together the power of LLMs with the Microsoft Graph to introduce enhanced, AI-based functionality into Microsofts productivity offering. For example, Copilot for Microsoft 365 within Microsoft Word, Excel and PowerPoint provide users with a way to create and analyze content in natrual language while leveraging information they have access to through their existing Microsoft 365 services such as SharePoint Online, OneDrive and Microsoft Teams. Additionally, Copilot for Microsoft 365 within Outlook and Teams can help people to organize their day, draft communications and stay on top of their work with easy summerization and recap functionality.

In this article, I won’t be explaining the many use-cases for Copilot for Microsoft 365 – thats been done a million times over elsewhere. Instead, I want to focus on one of the most common challenges that I see with my customers in understanding and deploying Copilot for Microsoft 365, data permissions.

One of the main selling points for Copilot for Microsoft 365 is that by connecting with the Microsoft Graph, it can access your data to provide insights and context based on what you as a user have access to. This statement alone usually raises some questions and concerns within organizations around data security so I will clear up some of those frequently asked questions here:

• Copilot for Microsoft 365 will only provide responses based on content a user already has access to. It will not return all the data from the restrricted Payroll site for users who don’t have access to it.
• Content created by Copilot for Microsoft 365 is generated with built-in copyright protections (see Microsofts official statement on this here)
• Copilot for Microsoft 365 is aware of Sensitivity Labelling and if it generates content from a labelled file, the output will also be labelled.
• Copilot for Microsoft 365 has built-in protections around sensitive topics such as revealling passwords or salaries (See Figure 1)

In this article, I want to focus on the data access question. Copilot for Microsoft 365 will not reveal data that a user doesn’t have access to as I mentioned above, but a common issue is that we don’t generally know what a user has access to – particularly in larger environments. There is a risk that if data permissions are not correct, any exposed data will become more visible through Copilot for Microsoft 365.

So what data will Copilot access on behalf of a user?

Copilot for Microsoft 365 will return data from your organization which the user explicitly has access to. This means that if a user has direct permissions on a file, or is a member of a specific site, that data can be returned. Where it gets a bit more confusing, is with sharing links. Generating a sharing link can be done using a number of options:

• Share with Anyone
• Share with People in your Organization
• Share with People with Existing Access
• Share with Specific People

For each of these type of sharing links, the expectation from Copilot for Microsoft 365 to expose data is a little bit different.

When a “Anyone” or “Anonymous” link is generated, anyone (internal or external) can access the file as long as they have the link. These links do not require authentication and I generally recommend thinking long and hard before allowing these type of links within your tenant. As they don’t authenticate the user, they also don’t modify the access for the file or folder, meaning that files ahred through “Anyone” links will not be surfaced through Copilot for Microsoft 365.

When a link is created using the “People in your Organization” option, a specific authenticated link is created. If a user does not open the link, they don’t get access to the file – meaning Copilot for Microsoft 365 will not be able to access this file. If a user clicks the link however, they will then be added to the access list. Once they are present on the access list, Copilot for Microsoft 365 will be able to return data from this file.

When a link is shared with “People with existiing access”, users are already present on the access list so the data is already accessible through Copilot for Microsoft 365.

When a link is shared with “Specific people”, users get added to the access list at the time of sharing, making the data accessible to any users shared using this method.

With all these different methods of people gaining access to files, it can be difficult to manage and understand what content will be returned by Copilot for Microsoft and from where. Many organizations I speak to have concerns about just what Copilot will access and respond with. The key point here, is that Copilot for Microsoft 365 will not return data that a user does not already have access to, but we first need to understand what that access looks like. To help with this, I have built a report which provides insights into data access across the tenant to help organizations understand what their data permissions look like.

Check out my article on runnning a Copilot for Microsoft 365 Data Permissions Assessment here!