Microsoft is rolling out a fix for a newly disclosed issue in its Copilot AI assistant that allowed the tool to read and summarize emails marked “confidential” in Outlook. The bug affects Copilot Chat within Microsoft 365 accounts and was confirmed in an internal report cited by security outlet BleepingComputer.
According to the report, Copilot could access and summarize emails stored in the Sent and Drafts folders even when those messages carried confidentiality labels. Such labels are typically used to prevent automated tools and unauthorized parties from accessing sensitive information, including legal correspondence, contracts, medical data, and government communications.
The issue, tracked internally under a Microsoft bug identifier, is considered serious because it undermines safeguards intended to protect sensitive communications. Microsoft says a fix is being deployed, though the company has not provided a clear timeline for when all affected accounts will receive it. The full incident report is only visible to Microsoft 365 administrators, limiting public details on the scope of the problem.
Microsoft has not disclosed how many users or organizations were impacted but acknowledged that the “scope of impact may change” as its investigation continues. The company also has not indicated whether any of the data accessed by Copilot could have been used to train AI systems or stored beyond the immediate summarization function.
The incident highlights ongoing concerns about how AI assistants interact with sensitive workplace data. As companies integrate AI tools more deeply into email and document workflows, ensuring that confidentiality labels and other protections function as intended remains a critical challenge.
