Microsoft Patches Copilot Flaw That Exposed Confidential Emails

A security vulnerability in Microsoft's Copilot for Microsoft 365 allowed unauthorized users to view confidential emails, highlighting persistent risks as AI tools become workplace staples. The flaw, now patched, let employees query the AI assistant and receive summaries or excerpts from emails marked with restrictive sensitivity labels, bypassing Microsoft's own Purview information protection system.

TechRepublic first reported the issue, noting that Copilot was pulling from emails with labels like 'Confidential' and presenting that information to users without proper clearance. Microsoft has resolved the problem, stating Copilot now correctly respects these labels. The company did not specify how many organizations were affected or for how long the vulnerability existed.

This incident underscores a fundamental challenge with integrated AI assistants: their design to be comprehensively helpful can conflict with strict data governance. The systems search across vast organizational data to generate answers, but if permission boundaries aren't perfectly enforced at every stage, sensitive information leaks through.

Security analysts point to a common underlying issue: many organizations have poorly maintained permission structures within their Microsoft 365 environments. What was once a latent risk—'permission sprawl' where files are accessible to too many users—becomes active when an AI can surface that data unprompted.

The event serves as a practical reminder for organizations using or evaluating such tools. Actions should include verifying the patch is active, conducting thorough audits of data permissions across SharePoint, OneDrive, and Exchange, and establishing clear policies on which data repositories AI tools can access. As AI assistants from multiple vendors embed deeper into business workflows, ensuring governance frameworks keep pace with deployment speed is no longer optional.

Source: Webpronews