Microsoft Says it is Not Training Copilot AI on Your Microsoft 365 Data (Updated)

UPDATE: I added a quote from Microsoft after publishing the article. –Paul

This week, Microsoft refuted reports claiming that it is training its Copilot AI on data customers store in Microsoft 365. Which seems unnecessary, since it has said repeatedly that it will not do so, and it has a services agreement and privacy policy that govern this behavior.

Welcome to the new age of misinformation.

“Microsoft Office, like many companies [sic] in recent months, has slyly turned on an ‘opt-out’ feature that scrapes your Word and Excel documents to train its internal AI systems,” Nixcraft claimed on Twitter/X recently. “This setting is turned on by default, and you have to manually uncheck a box in order to opt out. If you are a writer who uses MS Word to write any proprietary content (blog posts, novels, or any work you intend to protect with copyright and/or sell), you’re going to want to turn this feature off immediately.”

Let’s ignore for a moment the hypocrisy of Nixcraft scraping a post from somewhere without attribution to make this claim and then not fact-checking it. Because they continue.

“Microsoft Connected Experiences means they collect your Word and Excel files for AI training,” Nixcraft continues. “Why not call it AI training? Why use such words? This is an unethical practice followed by a trillion-dollar corporation. How is this even legal? I am so glad that I don’t have Windows OS or MS office.”

And I’m so glad you don’t understand what this means but still have very strong opinions about it. Microsoft Connected Experiences has nothing to do with AI training. As Microsoft itself was forced to explain.

“We do not use customer data in the Microsoft 365 apps to train LLMs,” the Microsoft 365 team responded. “This setting only enables features requiring internet access like co-authoring a document.”

“These claims are untrue,” a Microsoft representative told me. “Microsoft does not use customer data from Microsoft 365 consumer and commercial applications to train foundational large language models. In certain instances, customers may consent to the use of their data for specific purposes, such as custom model development explicitly requested by some commercial customers.”

Microsoft’s X-based reply also includes a link to the Connected Experiences in Office page on Microsoft Learn, which further describes this functionality. As it notes, the primary use case here is document collaboration, but there are connected experiences that analyze your content to provide design, editing, and other recommendations, and other innocuous uses. They can be managed via corporate policy if you’re in a business, and many of them are optional.

Microsoft does use data from Bing, MSN, Copilot, and advertising interactions to train AI. If you use Copilot, that applies only to the chat experience. And if you use Copilot Pro, the paid subscription offering that works with Microsoft Office apps like Word and Excel, it doesn’t apply at all. Regardless, Microsoft allows consumers to opt out of Copilot AI training, and that’s true whether they’re using the free service or paying for Copilot Pro. (And it’s not at all convoluted.)

Microsoft communications lead Frank Shaw provided a similar response to an ill-informed wannabe AI expert on Bluesky. But as I often point out, when there are so many valid complaints to make, there’s no point in making any up.

“The Connected Experiences setting in Microsoft 365 people are referencing has no connection to how Microsoft trains foundational large language models,” Microsoft added. “This setting enables cloud-backed features many people have come to expect from productivity tools such as real-time co-authoring, cloud storage and tools like Editor in Word that provide spelling and grammar suggestions. The setting is not new and has been available since 2019.”