MicrosoftI am working on a solution using various MIP/ Purview features in an organisation of over 7000 user most of which have E5 licences.
the trainable classifiers are on and have a status of ready to use.
across the thousands of SPO sites, communication compliance policies, client side auto labelling policies for sensitivity labels.
i have a total of 5 detections appearing in content explorer all of which are so far from accurate.
ny concerns are 1) how are there not many more thousandths of detections, and 2) for the very few that they system detected almost all where detected as threat. They where not even close, I mean they were promotional emails, one had some Chinese characters when translated just said “no thanks”.
i have found when used in communication compliance the built in harassment / offensive images policy detects smiley face emojis, which are hardly inappropriate.
