The Multi-Agent Verification System (MAVS) Framework aims to improve fake news detection by leveraging a multi-agent system that enhances decision-making through multidimensional evaluation, mitigating adversarial attack vulnerabilities. MAVS utilizes four specialized agents (a GNN model and Generative AI models for fact-checking, stance-checking, and sentiment analysis) each operating independently and in parallel. The final classification is determined through a weighted aggregation of the agents’ outputs, optimized using Stochastic Gradient Descent (SGD)-based Logistic Regression to ensure optimal weight distribution. MAVS achieves an accuracy of 97.6% and an F1 score of 98%. Under a Multi-Agent Reinforcement Learning (MARL) attack, the system’s accuracy drops to 74.19% and the F1 score to 71%, while maintaining a precision of 100%. This highlights the framework’s resilience and ability to maintain high precision despite adversarial conditions. The proposed framework strengthens fake news detection by combining multiple verification strategies, reducing susceptibility to adversarial attacks. Future work includes refining agent interactions and exploring real-time deployment for broader applicability.