The company plans on using machine learning to try and spot bullying that might be taking place in photos and its captions. The use of machine learning doesn’t mean that it will automatically take such posts down, but rather it will be more proactive in flagging them so that human moderators can pick up on it quicker, where such posts could get taken down a lot faster compared to before, where moderators might need to wait for other people to flag it first.
This is slightly different from how Facebook deals with inappropriate photos, where Facebook had actually launched a somewhat controversial revenge porn prevention tool which required users to share photos they think might be posted, and where if those photos were to be uploaded, Facebook could detect them and prevent them from being uploaded entirely.
In addition to trying to prevent bullying in photos, Instagram will also be trying to crack down on abuse during broadcasts with optional comment filters for live videos.