Microsoft has addressed mounting concerns over claims that customer data from its Microsoft 365 apps, including Word and Excel, is being used to train its artificial intelligence models. The tech giant has categorically denied the allegations, which surfaced due to a misunderstanding about its privacy settings and user agreements.
Privacy Setting Misinterpreted
The confusion stemmed from a setting in Microsoft Office labeled “optional connected experiences.” This feature, designed to enhance user functionality—such as searching for online images or finding web-based information—was misconstrued as permission for AI model training.
This setting is toggled on by default, but it does not mention AI training in its disclosure. Additionally, a Microsoft learning document released on October 21, 2024, described several connected experiences in Office that “analyze your content” but did not explicitly clarify that customer data would not be used for AI model training, adding to the ambiguity.
Microsoft’s Official Response
In a statement via its Microsoft 365 X (formerly Twitter) account, the company clarified: “In the M365 apps, we do not use customer data to train LLMs [large language models]. This setting only enables features requiring internet access like co-authoring a document.”
Microsoft’s communications head, Frank Shaw, reiterated this stance on Bluesky, stating that the claims were false and emphasizing the company’s commitment to transparency.
Rising User Concerns Over AI Data Use
The incident highlights growing user anxiety about how tech companies handle personal data, particularly in relation to AI training. Earlier this year, Adobe faced a similar backlash when its terms of service were misinterpreted to suggest user-created content was being used to train generative AI systems. Adobe quickly updated its terms to clarify that this was not the case.
This unease is fueled by the practices of other tech companies, including Meta, X (formerly Twitter), and Google, which opt users into AI training by default, often without explicit user consent. The vast amounts of online content being scraped for AI training purposes have also intensified scrutiny.
