When Microsoft released Copilot in early 2023, it marked the start of a journey to integrate AI into everything Microsoft does, including in its key products Microsoft Office and Microsoft 365. Practically every software Microsoft now provides has a Copilot AI capability and is helping businesses to work more effectively. But, in a similar story of business’ transformation journeys to the cloud, Microsoft and all AI companies still have a responsibility to address privacy concerns and provide assurance that your AI interactions and data remain secure.

Pioneer of keeping confidential data safe

Microsoft Copilot has been the go-to AI assistant for business as it proactively launched early on with “Commercial Data Protection” to ensure the security of digital information and protect user privacy. Many other AI products at the time, and still do now, happily took every bit of data you put into it, regardless of how sensitive this data was, to further improve their products. In contrast, Microsoft took a more business and compliance-friendly approach:

• Microsoft would not store any interactions in or out of Copilot
• Microsoft would not know who was interacting with Copilot
• Microsoft does not use any of your data to train future models
• Microsoft does not mix Copilot used for business with the home version

This approach meant that there was a high level of confidence Microsoft Copilot was a safe option for financial services firms who wanted to use Generative AI but could not compromise the security or privacy of their data.

Eagle-eyed readers would note that the first two points are now in the past tense. As ever with AI, the pace of change is extremely fast and as of last month, this approach by Microsoft has now changed.

Where are your Copilot conversations stored?

The downside to not storing any Copilot interactions is governance. The lack of storing any interactions people in a firm have with Copilot meant it was impossible for a firm to see how people have been using this AI tool. This prevented the ability to place any technology guardrails that govern its use (outside of an AI User Policy).

As of last month, Microsoft has moved away from its “Commercial Data Protection” doctrine to “Enterprise Data Protection”. This change has meant Microsoft now stores any interactions people in your firm have with Copilot, and these interactions are associated with the person who made them. This is a more positive change that improves the overall security and privacy position.

Copilot interactions are stored within a person’s email mailbox within Microsoft 365, hidden away in Microsoft Outlook so you can’t actually see it. By storing interactions in a person’s mailbox, your Copilot interactions hold the same security and privacy benefits as the rest of your emails in Microsoft 365. This means Microsoft has the capability of storing them securely, but Microsoft can’t see them. Microsoft calls this the trust boundary, where everything Copilot-related sits within the same area as your emails, your OneDrive documents, SharePoint and Teams messages.

With this move of Copilot interactions to mailboxes, Microsoft has now begun to introduce the auditing and compliance capabilities that had been previously lacking, preventing some firms from even entertaining AI. The capabilities are similar to the same tools provided to email;

• Auditing of Copilot interactions
• eDiscovery of Copilot interactions
• Retention policies for Copilot chats

This change marks an alignment of Microsoft Copilot to the functionality demanded by financial services so that its use can be secured and private but also accountable.