Amazon Rekognition is an AWS service that other businesses license to automatically scan images and videos for inappropriate content — including explicit material, violence, hate symbols, drugs, and gambling — returning confidence-scored labels that allow platforms to flag unsafe material at scale before human moderators review it.

Amazon Rekognition Content Moderation uses pre-trained deep learning models exposed via API to detect inappropriate, unwanted, or offensive content across a detailed taxonomy of categories including explicit nudity, suggestive content, violence, drugs, tobacco, alcohol, hate symbols, gambling, and disturbing content. It returns each detected category with a confidence score. Customers can set confidence thresholds to determine what is automatically removed versus flagged for human review; AWS states human moderators typically need to review only 1–5% of total content volume after Rekognition pre-filtering. In February 2024, AWS launched an enhanced model adding 26 new moderation labels, expanding the taxonomy to three tiers, and adding detection of animated and illustrated content. The service is a commercial API sold to third-party developers and platforms; customers include Dream11 (100 million users), SmugMug/Flickr, and CoStar Group.