Microsoft's MAI models were trained using unlicensed web data, contradicting its claims of utilizing only clean and commercially licensed data. This practice aligns Microsoft with other AI firms that rely on fair use, placing the onus on website owners to restrict data access.