This paper addresses the challenge of explainable fake news detection by proposing a novel defense-based framework that leverages competing wisdom from raw reports. The framework consists of three main components: an evidence extraction module, a prompt-based reasoning module, and a defense-based inference module. The evidence extraction module identifies salient evidence from raw reports to split the wisdom into two competing parties. The prompt-based reasoning module uses a large language model (LLM) to generate justifications for both possible veracities. The defense-based inference module determines the veracity of a claim by modeling the defense among these justifications. Extensive experiments on two real-world datasets demonstrate that the proposed method outperforms state-of-the-art baselines in terms of fake news detection and provides high-quality justifications. The framework effectively mitigates the majority bias problem and enhances the explainability of fake news detection.This paper addresses the challenge of explainable fake news detection by proposing a novel defense-based framework that leverages competing wisdom from raw reports. The framework consists of three main components: an evidence extraction module, a prompt-based reasoning module, and a defense-based inference module. The evidence extraction module identifies salient evidence from raw reports to split the wisdom into two competing parties. The prompt-based reasoning module uses a large language model (LLM) to generate justifications for both possible veracities. The defense-based inference module determines the veracity of a claim by modeling the defense among these justifications. Extensive experiments on two real-world datasets demonstrate that the proposed method outperforms state-of-the-art baselines in terms of fake news detection and provides high-quality justifications. The framework effectively mitigates the majority bias problem and enhances the explainability of fake news detection.