Forensic Detection Of Toxic Chatbots, Deepfakes, And Automated Harmful Interactions: A State-of-the-art Review

Abstract

The Rapid Expansion Of Generative AI (GenAI) Technologies Has Enabled Unprecedented Synthesised Media And Sophisticated Automated Systems. The Existence Of Advanced Deepfake Technology, Generative Large Language Models (LLMs) That Can Produce Toxic Content, And Automated AI Botnets Has Created Substantial New Obstacles In Digital Forensics. The Present Review Consolidates Recent Work (2024–2025) Concerning The Forensics Of Three Colliding AI Threats: Audio-visual Deepfakes, Toxic Or Weaponised Chatbots, And The Automated Facilitation Of Harm. The Review Reconstructs The Shift From Artefact Detection Systems Toward The Use Of Behavioural, Semantic, And Multimodal Forensics. The Emerging Methodologies, Such As Biological Signal Processing, Audiovisual Correlation, LLM Systematisation Fingerprinting, And Traffic Encryption Biometrics, As Well As The Comprehensive Integration Of These Methodologies, Reveal Vital Yet Unrefined Methodologies That Focus On The Challenges Of Adversarial Attacks, Laundering Via Paraphrasing, And Performance Drops On Real-world Data. The Review Articulates The Necessity Of Well-founded, Transparent, And Standardizable Forensic Systems That Are Meant To Work In Adversarial Conditions.

Keywords