Microsoft has successfully fixed a machine learning model error that incorrectly flagged legitimate Gmail emails as spam in Exchange Online. The issue, tracked as EX1064599, began affecting users on April 25 and automatically moved legitimate emails to junk folders.

“Our machine learning model, which safeguards Exchange Online against risky email messages, incorrectly identified legitimate email messages as spam due to their similarity to email messages used in spam attacks,” Microsoft explained when acknowledging the problem.

By May 1, Microsoft confirmed they had reverted to the previous working version of the ML model, effectively resolving the false positive issue. During the disruption, administrators and users could create custom allow rules to prevent Gmail messages from being sent to junk folders.

“After monitoring, we’ve confirmed through our service health telemetry that reverting to the previous ML model has successfully remediated the impact,” Microsoft stated. “We’re continuing to investigate opportunities to improve our ML detection process to reduce false positive detections and prevent similar future impact.”

This isn’t the first spam filtering issue Microsoft has encountered this year. Recently, the company addressed another machine learning problem that incorrectly flagged Adobe emails as spam. Similar incidents occurred in March when anti-spam systems wrongly quarantined some users’ emails, and in October 2023 when a faulty anti-spam rule caused Microsoft 365 admins’ inboxes to be flooded with BCC copies of outbound emails mistakenly identified as spam.