The blistering integration of Internet of Things (IoT) technologies in health care and environmental monitoring has brought with it serious cybersecurity issues which have especially involved the safekeeping of sensitive and real-time data. The traditional centralized security architectures also suffer since they are vulnerable to single points of failure and scale poorly as there are more and more distributed IoT devices. To solve these problems, the paper would suggest an intelligent approach with the name Adaptive AI-Driven Federated Cyber Defense Framework (AA-FCDF). The model would merge federated learning with anomaly detection using reinforcement learning to provide real time decentralized real time threat identification without degrading privacy of data. Each IoT object has a lightweight local model that is trained on device-specific data and make encrypted model updates in a centralized aggregator periodically. This will facilitate the continued learning throughout the network and not compromise the information on health and the environment. Further, a scoring system based on trust performs a dynamic evaluation of the behaviour of the devices isolating or applying enhanced encryption protocols to potentially compromised nodes. The responsiveness to the two zero-day attack, internal attack, and spoofing issue is possible due to the flexibility of the system. Experiments on the simulation are showing higher accuracy of detection, less leakage of data, higher response speed than traditional models. The suggested AA-FCDF technique attained an overall accuracy of 97.4% in the assessment of cybersecurity efficiency inside IoT systems.