Abstract

The rapid proliferation of social media and digital communication has produced significant outcomes but it has also introduced major threats like cyberbullying, online harassment, offensive language, deepfake abuse, cyberstalking and online grooming. These threats impact countless users especially teenagers and women. Addressing them without automation is simply impractical given the massive quantity of content uploaded every second. This survey gathers outcomes from over 53 reviewed studies published between 2012 and 2026 to provide a coherent and comprehensive summary of how artificial intelligence (AI), machine learning (ML), deep learning (DL) and digital forensics are being used to detect and prevent these online threats. We examine all methods from traditional text classifiers to advanced transformer models like BERT and RoBERTa, from uni-modal text detection to multi-modal systems that analyze text, images, audio and video together. Focus is given to secure detection, multilingual-aware systems for non-English languages and explainable AI frameworks.