Blockchain-Based Trusted Data Sharing Framework for Secure and Auditable AI Model Training

AI models are highly effective and reliable based on integrity, provenance, and controlled usage of training data. The use of data sharing between several stakeholders in collaborative and distributed settings poses risks of unauthorized access, data tampering, and data poisoning that undermine trust and accountability in AI systems. To overcome these obstacles, this study introduces a blockchain-based trusted data sharing framework for secure and auditable AI model training pipelines. The solution consists of using a permissioned blockchain to store irreversible metadata of the datasets, cryptographic hashes, ownership, and access policy, and place the actual data off-chain to achieve scalability. Smart contracts provide a mechanism to automate registration of datasets, provide access control on a smart and intelligent basis, and record data access and data modification events to provide verifiable data provenance and end-to-end traceability. Before model training, the integrity of the data is automatically checked against the records on the chain to ascertain the use of authorized and undamaged data. Experimental evaluation in a simulated permissioned-blockchain environment indicates that the framework, applied as a practical design heuristic, can enhance data transparency, integrity, and auditability with reasonable performance overhead. The proposed solution helps to build credible and responsible AI, especially in data-sensitive environments, by ensuring that models are trained only on verified and authorized data.

@article{Amin2026,
  title     = {Blockchain-Based Trusted Data Sharing Framework for Secure and Auditable AI Model Training},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Sherif Amin},
  doi       = {10.14569/IJACSA.2026.0170748},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170748}
}