An Agentic AI Framework for Multi-Modal Fake Social Media Profile Detection

DOI

10.2991/978-94-6239-723-1_11How to use a DOI?

Keywords

Fake profile detection; multi-agent systems; intent classification; evidence fusion; social media forensics; explainable AI; large language model

Abstract

The rapid proliferation of generative artificial intelligence has fundamentally transformed the threat landscape of online social platforms, enabling adversaries to fabricate convincing fake profiles through synthetic face generation, automated text production, and coordinated inauthentic behaviour at scale. Conventional detection systems examine visual, textual, or behavioural signals in isolation through static rule-based pipelines, making them inherently fragile against accounts that manipulate one modality while appearing plausible in others. This paper proposes a multi-agent agentic solution for fake social media profile detection that treats the problem as a coordinated, intent-driven analytical task. The framework activates four specialised autonomous agents in parallel: a Vision Agent performing GAN-face detection through DCT frequency analysis, an NLP Agent combining rule-based scam pattern recognition with GPT-4o semantic reasoning, a Behavioural Agent applying Z-score profiling and Isolation Forest on posting cadence and engagement ratios, and a Social Graph Agent computing Sybil-indicative structural metrics using NetworkX. An Intent Classification Agent first interprets the analyst’s query classifying it as bot detection, impersonation, scam, or general audit and dynamically redistributes agent weights before orchestrating parallel execution. With a gate at U > 0.35 that overrides verdicts requiring human review. A dedicated Explainability component produces structured JSON audit logs compliant with EU AI Act Article 13 alongside plain-language moderator reports. Evaluated on TwiBot-20 (229,573 accounts) and an Instagram benchmark (12,628 accounts), the proposed solution achieves F1-scores of 0.913 and 0.891 with AUC-ROC values of 0.961 and 0.944 respectively, at a mean inference latency of 1,820 ms per profile.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Deepali Nikam
AU  - Archana Renushe
AU  - Shashank Joshi
PY  - 2026
DA  - 2026/07/14
TI  - An Agentic AI Framework for Multi-Modal Fake Social Media Profile Detection
BT  - Proceedings of the International Conference on Responsible, Risk-aware, and Regulated AI (RRRAI 2026)
PB  - Atlantis Press
SP  - 112
EP  - 126
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-723-1_11
DO  - 10.2991/978-94-6239-723-1_11
ID  - Nikam2026
ER  -