Microsoft has solved a problem with a machine learning model, which accidentally marks email from Gmail accounts as spam on the online exchange.
Tracked EX1064599 At the Microsoft 365 Admin Center, the issue began to influence users on 09:24 UTC on 25 April, automatically the email was wrongly tagged as malicious for junk folders.
“We have recognized that our machine learning (ML) models, which exchange the risky email messages online, are identifying legitimate email messages in the form of spam due to their similarity for email messages used in incorrectly spam attacks, which resulted in an impact,” the company accepted the ML model bug.
In a final update of the incident report added on UTC on 16:31 May 1, Microsoft said that it successfully returned the buggy ML model to the previous working version, which reduces the false positive issue. It has also been stated that the admin and users may be able to make custom permission rules to ensure that the Gmail messages were not sent to the junk folder, while the service was affected.
“After a period of surveillance, we have confirmed through our service health telemetry that the successful impact has been reduced to the previous ML model to resume,” Microsoft said. “We are continuing to check opportunities to improve our ML detection process to reduce false positive detection and prevent the same effect as future.”
While Redmund has not yet shared which areas or how many customers were affected, the issue of this service was tagged as an event, which usually involves noticeable user effects.
Microsoft has handled similar issues since the beginning of the year, causing the email to have been incorrectly tagged as spam or quarantine. For example, last week, the company had reduced another machine-learning issue, which accidentally marked Adobe Email as online spam on the exchange.
In March, it addressed another exchange online Phals positive, making the anti-spam systems wrongly quarant the emails of some users.
In October 2023, it was also to disable a poor anti-spam rule, which was misinterpreted as spam with outbound emails with blind carbon copies (BCC) with inbox of Microsoft 365 Admins, while in August 2024, in August 2024, it reduced the untouchable and self-tagging emails in August 2024, while in August 2024 It was framed in the form.
Based on the analysis of 14M malicious tasks, search for the top 10 MITERAT & CK techniques behind the 93% attacks and how to defend them against them.
