Microsoft Denies Training AI on Customer Data Following Uproar

Microsoft Corp. rejected claims that it uses customer data from Microsoft 365 consumer and commercial applications, such as Word and Excel, to train large language models, after user worries spread on social media.

Allegations emerged a few days ago after some users noticed they were automatically opted into Microsoft 365’s Connected Experiences, a 2019 setting that allows product features to be connected to the internet and the cloud.

The claims spread quickly among Microsoft users over what they perceived as a “sneaky” update to the provider’s terms of use, which they said could allow the company to scrape Microsoft Office documents to train AI models.

The setting in question didn’t specifically refer to AI training.

“These claims are untrue,” a Microsoft spokesperson told Bloomberg Law Wednesday, noting that some customers may consent to their data being used in “certain instances,” such as for custom model development when it is “explicitly requested.”

The Connected Experiences setting being referenced has “no connection to how Microsoft trains foundational large language models,” the spokesperson said. This setting enables cloud-backed features such as real-time co-authoring, cloud storage, and tools like Editor in Word that provide spelling and grammar suggestions, they said.

Users of online platforms have grown increasingly worried of their data—whether it be social media post, contracts, or pictures—being used to train organizations’ latest AI models.

This summer, Adobe Systems Inc. faced similar uproar when it updated its terms of use, which users claimed allowed Adobe to seize their intellectual property and use their data to feed AI models. The Photoshop and Illustrator maker issued updated terms of use days later, in which it specifically stated the company doesn’t train its Firefly AI models on its cloud content.