<p>An Internet of Things-enabled smart healthcare system provides continuous monitoring to enhance patient care and their treatment outcomes. This system involves patient-sensitive data and their health records; thus, it is affected by various potential threats, privacy breaches, and unauthorized activities. Several blockchain-based techniques have been developed by researchers to detect potential threats and enhance the privacy of patient data, but they are limited by low scalability, poor detection accuracy, poor adaptability, and high complexity issues. This research work designs a blockchain-based federated learning framework named AI-HealthChain for improving privacy and data security. The AI-HealthChain framework adopts a federated learning mechanism for decentralized model training, where client privacy data are kept on the local model, and only model updates are shared. For local model training, an autoencoder model is implemented that effectively detects various security attacks. The convolutional neural network-based hashing mechanism is developed by this study for generating a hash value to optimize cryptographic security, thereby improving security and data integrity. To improve energy efficiency and scalability, the proposed AI-HealthChain framework designs an adaptive proof of utility consensus mechanism in the blockchain layer that prioritizes high-utility nodes and validates each client's updates. A natural language processing-driven smart contract is used to modify natural language agreements into a smart contract to manage patient consent efficiently. Experimental evaluation on healthcare datasets indicates that the proposed AI-HealthChain framework attains a lower latency of 250&#xa0;ms, throughput of 9733 transactions per second, 99.6% entropy, 0.008 collision resistance, and execution time of 15.6&#xa0;ms.</p>