Microsoft has confirmed that a software bug allowed its Copilot AI to summarise customers’ confidential emails for several weeks without their approval.
The issue, first reported by BleepingComputer, enabled Copilot Chat to access and generate summaries of email content dating back to January, even for organisations with data loss prevention policies in place to prevent sensitive information from being ingested into Microsoft’s large language model.
Copilot Chat is available to paying Microsoft 365 customers and provides an AI-powered chat experience across Office applications, including Word, Excel, and PowerPoint.
Microsoft said the bug — which administrators can track under the reference number CW1226324 — resulted in draft and sent email messages “with a confidential label applied” being “incorrectly processed by Microsoft 365 Copilot chat.”
The company said it began rolling out a fix earlier in February. A Microsoft spokesperson did not respond to a request for comment, including questions about how many customers may have been impacted by the issue.
Earlier this week, the European Parliament’s IT department told lawmakers that it had blocked built-in AI features on work-issued devices, pointing to concerns that AI tools could upload potentially confidential correspondence to the cloud.
