Microsoft said it does not use Microsoft 365 user data to train large language models (LLMs) that become the basis of AI algorithms. This clarification concerns messages that have circulated on the Internet in recent weeks, the authors of which claim that the company has activated a function for collecting the contents of Word and Excel documents and users must disable it themselves so that this data is not used to train neural networks.
According to a Microsoft spokesperson, the confusion stemmed from an option in the Privacy Options menu called More Connected Experiences. He noted that this function allows you to “search for information on the Internet” and it is indeed activated by default, but its description makes no mention of AI training. It appears the confusion may have stemmed from a Microsoft training document published in September that outlined a long list of connected Office capabilities that analyze user content. It did not say that office suite user content was being used for LLM training.
«In Microsoft 365 Apps, we do not use customer data for LLM training. This setting only enables features that require internet access, such as collaborative document editing,” reads a post on the Microsoft 365 social media account X. Microsoft head of communications Frank Shaw also denied in his Bluesky account claims that the software giant is using user data from Microsoft 365 to train neural networks.