Microsoft confirms it doesn't use your Office docs to train AI

Microsoft is one of the tech giants running the ongoing AI race aggressively. But, that doesn’t mean it can’t be the topic of controversy. Allegedly, Microsoft Office quietly enabled a feature letting it scrap data from Word and Excel documents to train its AI.

Microsoft clarifies it doesn’t train its AI on your Office docs

Microsoft busted that rumor saying it doesn’t train AI with the help of users’ Microsoft 365 apps’ data. The confusion apparently picked up the hype after a few users spotted a privacy setting in Microsoft Office. According to this post on X, there’s a toggle for “optional connected experiences.”

Interestingly, the toggle is enabled by default, and that seems to bother a lot of users. On top of that, there’s no mention of AI training in the disclosure. Microsoft says that this feature allows users to “search for online pictures” or “find information available online.”

That’s not all, a Microsoft learning document dated October 21, 2024, also contributed to this confusion. For those unaware, that document includes a long list of connected experiences in Office that “analyze your content.” Here, it’s worth noting that Microsoft hasn’t explicitly excluded AI training.

In response to debunk the claims, the Microsoft 365 X account posted “In the M365 apps, we do not use customer data to train LLMs. This setting only enables features requiring internet access like co-authoring a document.” In addition, Microsoft’s communications head, Frank Shaw, also took to his Bluesky account to inform users about the same. Not to forget, Bluesky also recently confirmed it doesn’t train AI on users’ posts.

Similar incidents from the past

To catch you up, Adobe and LinkedIn have faced similar criticism in the past. Many users misinterpreted Adobe’s user terms claiming that the company was training its generative AI (Gen-AI) on users’ creative work. Adobe then clarified the claims by updating the language of its terms of service.

The LinkedIn-related controversy started with a toggle in the privacy settings. However, the company’s VP and general counsel, Blake Lawit, then addressed the same hinting it to be an opt-out setting. All that said, incidents like these suggest the need for companies to obtain permission prior to training AI models on their data. If not, they should at least try to inform users in clear-cut words.