This paper proposes a defense-based explainable fake news detection framework that leverages the competing wisdom in raw reports to provide veracity justifications. The framework consists of three components: an evidence extraction module to split the wisdom into two competing parties and detect salient evidences, a prompt-based reasoning module that uses a large language model (LLM) to generate justifications for two possible veracities, and a defense-based inference module that determines veracity by modeling the defense among these justifications. The framework is evaluated on two real-world benchmarks, RAWFC and LIAR-RAW, and outperforms state-of-the-art baselines in terms of fake news detection and provides high-quality justifications. The experiments show that the proposed method achieves state-of-the-art performance in both veracity prediction and explanation quality, demonstrating the effectiveness of the defense-based approach in mitigating majority bias and improving the explainability of fake news detection. The framework is able to effectively separate competing parties from the sea of wisdom, generate competing justifications, and determine the veracity of a claim through a defense-like process. The results show that the proposed method is robust to different LLMs and is not limited by the size of the LLM component. The framework also provides human-readable explanations that align with the predicted veracity label, making it a valuable tool for automated fake news detection.This paper proposes a defense-based explainable fake news detection framework that leverages the competing wisdom in raw reports to provide veracity justifications. The framework consists of three components: an evidence extraction module to split the wisdom into two competing parties and detect salient evidences, a prompt-based reasoning module that uses a large language model (LLM) to generate justifications for two possible veracities, and a defense-based inference module that determines veracity by modeling the defense among these justifications. The framework is evaluated on two real-world benchmarks, RAWFC and LIAR-RAW, and outperforms state-of-the-art baselines in terms of fake news detection and provides high-quality justifications. The experiments show that the proposed method achieves state-of-the-art performance in both veracity prediction and explanation quality, demonstrating the effectiveness of the defense-based approach in mitigating majority bias and improving the explainability of fake news detection. The framework is able to effectively separate competing parties from the sea of wisdom, generate competing justifications, and determine the veracity of a claim through a defense-like process. The results show that the proposed method is robust to different LLMs and is not limited by the size of the LLM component. The framework also provides human-readable explanations that align with the predicted veracity label, making it a valuable tool for automated fake news detection.