February 23, 2026, 4:35 pm | Read time: 2 minutes
Microsoft’s AI assistant Copilot can automatically summarize emails. This saves time and reduces the reading workload in daily work. However, the system must read the content of the messages. The data processing occurs on external cloud servers. This can be problematic, especially with sensitive content. This affects confidential business data as well as private information.
Microsoft therefore offers the option to mark certain emails as confidential. Copilot is then supposed to have no access to these messages. However, there was apparently an error here.
Technical Error in Work-Tab Chat
As reported by the website “Bleeping Computer” citing an entry in Microsoft’s administrator portal, Copilot was able to read and summarize emails marked as confidential in the Work-Tab Chat of the Microsoft 365 suite. Affected were messages in the “Sent Items” and “Drafts” folders.
A Work-Tab Chat is a workspace within Microsoft 365 where Copilot assists with tasks. According to the report, Copilot incorrectly processes emails with confidentiality markings. This means the protection function did not work as intended. Microsoft is currently informing affected companies about the issue. The company points out that the AI assistant incorrectly processed these emails.
Microsoft Causes Stir with Secret Word Filter
How to Avoid the Price Increase for Microsoft 365
Problem Apparently Known Since January
The error is apparently not entirely new. Initial reports of the phenomenon appeared as early as January 2026. According to its own statements, Microsoft has been working on a solution since early February. Whether the error has been completely resolved remains unclear. It is also unknown how many companies and individuals are actually affected. Microsoft states that developers are currently communicating with affected companies. The goal is to assess the effectiveness of measures already taken.
There is currently no specific timeline for a final solution. Microsoft will announce it as soon as it is available. This information is from February 11, 2026, and there have been no new updates since then.
Criticism of Extensive AI Access
The extensive access of modern AI assistants to data and computers regularly draws criticism. Privacy advocates warn that sensitive information could be inadvertently processed or shared.
At the same time, such systems create new potential attack surfaces. Attackers could try to penetrate systems via AI functions. The current incident involving Copilot highlights the importance of functioning protection mechanisms.