This paper proposes AKA-Fake, a novel Adaptive Knowledge-Aware Fake News Detection model for multimodal news detection. The model addresses the limitations of existing knowledge-enhanced detection methods, which often use static knowledge-context modeling and suffer from noisy and irrelevant knowledge. AKA-Fake generates an adaptive knowledge subgraph under a reinforcement learning paradigm, which consists of a subset of entities and contextual neighbors in the knowledge graph, restoring the most informative knowledge facts. A novel heterogeneous graph learning module is further proposed to capture the reliable cross-modality correlations via topology refinement and modality-attentive pooling. The model is evaluated on three popular datasets, and the experimental results demonstrate its superiority over existing methods. The key contributions include the first investigation of adaptive knowledge learning for multimodal fake news detection, the proposal of AKA-Fake to learn adaptive knowledge subgraphs and integrate knowledge with multimodal content, and extensive experiments showing consistent performance improvements. The model effectively captures multimodal content and assimilates reliable external knowledge, leading to superior overall performance.This paper proposes AKA-Fake, a novel Adaptive Knowledge-Aware Fake News Detection model for multimodal news detection. The model addresses the limitations of existing knowledge-enhanced detection methods, which often use static knowledge-context modeling and suffer from noisy and irrelevant knowledge. AKA-Fake generates an adaptive knowledge subgraph under a reinforcement learning paradigm, which consists of a subset of entities and contextual neighbors in the knowledge graph, restoring the most informative knowledge facts. A novel heterogeneous graph learning module is further proposed to capture the reliable cross-modality correlations via topology refinement and modality-attentive pooling. The model is evaluated on three popular datasets, and the experimental results demonstrate its superiority over existing methods. The key contributions include the first investigation of adaptive knowledge learning for multimodal fake news detection, the proposal of AKA-Fake to learn adaptive knowledge subgraphs and integrate knowledge with multimodal content, and extensive experiments showing consistent performance improvements. The model effectively captures multimodal content and assimilates reliable external knowledge, leading to superior overall performance.