Cognitive Inception: Agentic Reasoning against Visual Deceptions by Injecting Skepticism

2025-11-21

AI Summary

Multimodal large language models (MLLMs) struggle to distinguish AI-generated from authentic visual content, rendering their reasoning vulnerable to visual deception and compromising reliability. Method: We propose Inception—a novel, fully inference-based skeptical enhancement agent architecture. It introduces internal and external skeptical agents that collaboratively perform multi-agent iterative reasoning, emulating human cognitive doubt mechanisms without relying on assumptions about data distribution, training annotations, or generator-specific features. Contribution/Results: Inception enables self-consistent visual authenticity verification through purely reasoning-driven validation. It exhibits strong generalization across diverse generative models and data domains. Evaluated on the AEGIS benchmark, it achieves state-of-the-art performance, significantly outperforming the strongest existing baselines. To our knowledge, Inception is the first framework to systematically enhance MLLMs’ generalizable capability for authenticating AI-generated content (AIGC), advancing trustworthy multimodal reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Large Vision Models
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

As the development of AI-generated contents (AIGC), multi-modal Large Language Models (LLM) struggle to identify generated visual inputs from real ones. Such shortcoming causes vulnerability against visual deceptions, where the models are deceived by generated contents, and the reliability of reasoning processes is jeopardized. Therefore, facing rapidly emerging generative models and diverse data distribution, it is of vital importance to improve LLMs' generalizable reasoning to verify the authenticity of visual inputs against potential deceptions. Inspired by human cognitive processes, we discovered that LLMs exhibit tendency of over-trusting the visual inputs, while injecting skepticism could significantly improve the models visual cognitive capability against visual deceptions. Based on this discovery, we propose extbf{Inception}, a fully reasoning-based agentic reasoning framework to conduct generalizable authenticity verification by injecting skepticism, where LLMs' reasoning logic is iteratively enhanced between External Skeptic and Internal Skeptic agents. To the best of our knowledge, this is the first fully reasoning-based framework against AIGC visual deceptions. Our approach achieved a large margin of performance improvement over the strongest existing LLM baselines and SOTA performance on AEGIS benchmark.

Problem

• Improving LLMs' ability to verify visual input authenticity against AI-generated deceptions
• Addressing vulnerability of multimodal models to visual deception from generated content
• Enhancing generalizable reasoning through skepticism injection against AIGC visual manipulations

Innovation

• Injecting skepticism into multi-modal LLMs
• Agentic reasoning framework with iterative enhancement
• Fully reasoning-based approach against visual deceptions