A system enhanced with multi-agent search capabilities for the automated detection and classification of harmful online content.

Abstract

A multi-agent retrieval-enhanced system for the automated detection and classification of harmful online content, comprising: a content capture module configured to receive a content item from one or more open-source information sources; a knowledge base comprising a variety of categorized knowledge entries linked to threat indicators, coded language indicators, counter-speech indicators, legal threshold definitions, disinformation indicators, and multilingual reference information; a retrieval engine communicatively connected to the knowledge base and configured to retrieve one or more knowledge entries relevant to the content item based on a relevance determination process;a variety of AI agents configured to receive the content contribution and retrieved knowledge entries, the variety of AI agents including: a content classification agent configured to identify one or more threat categories associated with the content item; a counterspeech assessment agent configured to determine whether the content item is journalism, academic analysis, satire, community solidarity, condemnation of harmful activities, or other non-threatening content; a context and intent assessment agent configured to identify one or more context indicators associated with the content item, including intent, urgency, audience, target identification, or operational specificity;and a coordination agent configured to receive the results generated by the content classification agent, the counterspeech assessment agent, and the context and intent assessment agent, and to produce a consolidated threat assessment; wherein the consolidated threat assessment includes a threat level and a recommended action associated with the content item.