Reasoning-Aware AIGC Detection via Alignment and Reinforcement

arXiv cs.AI / 4/22/2026

📰 NewsDeveloper Stack & InfrastructureTools & Practical UsageModels & Research

共有:

Key Points

The paper introduces AIGC-text-bank, a multi-domain dataset designed to support more robust AI-generated text detection across diverse LLM sources and authorship scenarios.
It proposes REVEAL, a detection framework that first generates an interpretable reasoning chain and then performs classification based on it.
REVEAL is trained in two stages: supervised fine-tuning to develop reasoning ability, followed by reinforcement learning to boost detection accuracy, improve logical consistency, and reduce hallucinations.
Experimental results indicate that REVEAL reaches state-of-the-art performance on multiple benchmarks and is positioned as a more transparent approach to AIGC detection.
The project is released as open source via https://aka.ms/reveal.

Abstract

The rapid advancement and widespread adoption of Large Language Models (LLMs) have elevated the need for reliable AI-generated content (AIGC) detection, which remains challenging as models evolve. We introduce AIGC-text-bank, a comprehensive multi-domain dataset with diverse LLM sources and authorship scenarios, and propose REVEAL, a detection framework that generates interpretable reasoning chains before classification. Our approach uses a two-stage training strategy: supervised fine-tuning to establish reasoning capabilities, followed by reinforcement learning to improve accuracy, improve logical consistency, and reduce hallucinations. Extensive experiments show that REVEAL achieves state-of-the-art performance across multiple benchmarks, offering a robust and transparent solution for AIGC detection. The project is open-source at https://aka.ms/reveal