Microsoft has resolved a problem with a machine studying mannequin that mistakenly flagged emails from Gmail accounts as spam in Alternate On-line.
Tracked as EX1064599 within the Microsoft 365 admin middle, the difficulty began impacting customers on April 25 at 09:24 UTC, mechanically transferring emails erroneously tagged as malicious to the junk folder.
“We have recognized that our machine studying (ML) mannequin, which safeguards Alternate On-line in opposition to dangerous e mail messages, is incorrectly figuring out authentic e mail messages as spam as a consequence of their similarity to e mail messages utilized in spam assaults, which is leading to impression,” the corporate defined when it acknowledged the ML mannequin bug.
In a closing replace to the incident report added on Might 1 at 16:31 UTC, Microsoft stated it efficiently reverted the buggy ML mannequin to the earlier working model, mitigating the false optimistic challenge. It additionally added that admins and customers might have additionally been capable of create customized permit guidelines to make sure that Gmail messages weren’t despatched to the junk folder whereas the service was impacted.
“After a interval of monitoring, we have confirmed by way of our service well being telemetry that the completion of reverting to the earlier ML mannequin has efficiently remediated impression,” Microsoft added. “We’re persevering with to research alternatives to enhance our ML detection course of to cut back false optimistic detections and forestall related future impression.”
Whereas Redmond has but to share what areas or what number of clients had been affected, this service challenge was tagged as an incident, which usually entails noticeable person impression.
Microsoft has dealt with related points because the begin of the yr, resulting in emails being incorrectly tagged as spam or quarantined. As an example, final week, the corporate mitigated one other machine-learning challenge that mistakenly flagged Adobe emails in Alternate On-line as spam.
In March, it addressed one other Alternate On-line false optimistic, inflicting anti-spam techniques to quarantine some customers’ emails incorrectly.
In October 2023, it additionally needed to disable a foul anti-spam rule flooding Microsoft 365 admins’ inboxes with blind carbon copies (BCC) of outbound emails mistakenly flagged as spam, whereas in August 2024, it mitigated an Alternate On-line bug tagging emails containing photographs as malicious and mechanically quarantining them.
Primarily based on an evaluation of 14M malicious actions, uncover the highest 10 MITRE ATT&CK methods behind 93% of assaults and defend in opposition to them.
