| Artificial Intelligence (AI) systems are increasingly being deployed in production healthcare environments to support diagnosis, prognosis, and clinical decision-making. However, existing deployment pipelines provide limited guarantees of model integrity, provenance, and auditability once models are placed into operation, leaving high-stakes systems vulnerable to undetected tampering and misuse. In this paper, we propose a blockchain–based architecture called TwAIChain for secure, auditable, and tamper-resistant access to AI models in healthcare. The framework combines decentralised storage via the InterPlanetary File System (IPFS) with the immutability of an Ethereum-based blockchain to keep the model trustworthy. Cryptographic hashing and smart contracts enforce integrity checks and immutably log model access events in accordance with a defined threat model. We introduce a new set of trust and performance metrics and evaluate the system through algorithmic analysis and empirical measurements. The results demonstrate reliable detection of model tampering prior to inference while maintaining low latency and operational cost, supporting the practical deployment of trustworthy AI in regulated healthcare environments. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.