DecepGPT: Schema-Driven Deception Detection with Multicultural Datasets and Robust Multimodal Learning

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, long

Summary

DecepGPT introduces a schema-driven approach to multimodal deception detection, addressing limitations in existing benchmarks. The system constructs reasoning datasets by augmenting current benchmarks with structured cue-level descriptions and reasoning chains, enabling auditable reports. It also releases T4-Deception, the largest non-laboratory deception detection dataset with 1695 samples, covering identity pretense across four countries: the U.S., Germany, Vietnam, and Bulgaria, with an average segment duration of 3.65s. Furthermore, DecepGPT proposes two robust learning modules: Stabilized Individuality-Commonality Synergy (SICS) for refining multimodal representations and Distilled Modality Consistency (DMC) to prevent unimodal shortcut learning. Experiments on three established benchmarks and the T4-Deception dataset demonstrate state-of-the-art performance in in-domain, cross-domain, and cross-cultural scenarios.

Key takeaway

For AI Scientists and Machine Learning Engineers developing high-stakes deception detection systems, you should prioritize auditable outputs and cross-cultural robustness. Implement schema-constrained reasoning pipelines to provide verifiable evidence, moving beyond binary predictions. Consider integrating modules like SICS for robust feature refinement and DMC for mitigating unimodal shortcut learning, especially when working with limited or culturally diverse datasets like T4-Deception, to ensure reliable and explainable forensic analysis.

Key insights

Multimodal deception detection requires auditable reasoning and robust cross-cultural generalization, addressed by schema-driven data and specialized learning modules.

Principles

• Deception detection needs verifiable evidence, not just binary labels.
• Multicultural datasets improve generalization across diverse contexts.
• Robust learning modules mitigate small-data issues and unimodal shortcuts.

Method

DecepGPT uses a Human-in-the-Loop pipeline to generate structured reasoning datasets. It integrates SICS for polarity-aware feature refinement and DMC for consistency distillation, ensuring auditable reports from multimodal inputs.

In practice

• Augment existing datasets with cue-level descriptions for audit trails.
• Utilize SICS to stabilize multimodal feature representations.
• Apply DMC to prevent unimodal shortcut learning in MDD.

Topics

• Multimodal Deception Detection
• Auditable AI
• T4-Deception Dataset
• SICS Module
• DMC Regularizer
• Cross-Cultural AI

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Rift: A Conflict Signature for Deception in Language Models — Deception in language models leaves a detectable internal "conflict signature" distinct from honest error.
• Pressure-Testing Deception Probes in LLMs: Scaling, Robustness, and the Geometry of Deceptive Representations — Deception detection probes in LLMs fail due to training data narrowness, not architectural limits, but recover with style augmentation.
• CRED-1: Building an Open Domain Credibility Dataset for Misinformation Pre-Bunking — Multi-signal credibility scoring offers a nuanced, dynamic, and transparent alternative to binary domain labeling for misinformation pre-bunking.
• Zero-source LLM Hallucination Detection with Human-like Criteria Probing — HCPD detects LLM hallucinations in zero-source settings by emulating human multi-criteria reasoning for truthfulness, outperforming baselines.
• KITE: A Tri-Modal Transformer Integrating Text, Images, and Knowledge Graphs for Fake News Detection — KITE integrates text, images, and knowledge graphs via a tri-modal transformer for enhanced fake news detection and interpretability.
• Bhramastra at #SMM4H-HeaRD 2026: A Multi-Stage Hunter-Judge Pipeline using DSPy-Optimized LLMs for Multilingual ADE Detection — A decoupled Hunter-Judge architecture with DSPy-optimized LLMs effectively detects multilingual ADEs, achieving top performance and zero-shot generalization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.