Microsoft has resolved a problem in Exchange Online that wrongly flagged Gmail emails as spam, affecting users since April 25.
The issue was caused by a machine learning (ML) model used to protect against risky messages, which mistakenly tagged legitimate Gmail emails as malicious and moved them to users’ junk folders.
The problem, tracked as EX1064599, was due to the ML model confusing real emails with spam because of similarities in their content. Microsoft admitted the mistake and rolled back the ML model to its earlier, working version on May 1, which successfully fixed the issue.
“We’ve confirmed through our monitoring that reverting to the previous model has resolved the impact,” Microsoft said.
Admins had the option to create custom allow rules during the outage to prevent Gmail messages from going to spam. However, Microsoft didn’t specify which regions or how many users were affected.
