Microsoft fixes machine learning bug flagging Adobe emails as spam

Microsoft mitigated a false-positive in an Exchange Online ML model that flagged legitimate Adobe URLs and emails as malicious, causing alerting and delivery issues; Microsoft used Replay Time Travel and updated ML logic to reduce future false positives. As a consequence, MANY free-plan users and Defender XDR customers uploaded more than a thousand Adobe Acrobat files (containing sensitive corporate data from hundreds of companies) to the public ANY.RUN sandbox, prompting ANY.RUN to make analyses private while some confidential documents remained publicly shared.