Microsoft trained its MAI models on unlicensed web data despite promising "enterprise grade, clean and commercially licensed data"

Editor's take

Microsoft's MAI large language models were trained, in part, on publicly scraped web data, contradicting its public assertions of using exclusively "enterprise grade, clean and commercially licensed data."

This development is significant because Microsoft has positioned its AI offerings, particularly for enterprise clients, as a more ethically sourced and legally secure alternative to models trained on potentially problematic datasets. The discrepancy raises questions about the actual provenance of the data used to build these models, potentially impacting trust with businesses concerned about intellectual property and licensing risks associated with AI-generated content.

Future scrutiny will focus on the extent of this "partly" and whether Microsoft can provide a clearer audit trail for its training data. The company's ability to reassure enterprise customers about data integrity, especially in light of ongoing copyright lawsuits against other AI developers, will be crucial for maintaining its competitive edge in the cloud AI market.