<p>The health informatics field’s pursuit of personalized healthcare continuously faces constraints from patients, clinicians, and resource limitations. Recent advances in artificial intelligence (AI) and machine learning (ML) models have led to their widespread adoption in personalized genomic research for their outstanding predictive capabilities for drug responses to assist in personalized healthcare for tailored therapies and many other applications. Despite their growing use, such models often operate as black boxes, tempering, lacking sources to verify whether a prediction was generated honestly by a model’s input or manipulated post hoc. Over these challenges, this study presents a decentralized model that integrates AI predictive modeling with blockchain-based verification to ensure the integrity, traceability, trust, and reproducibility of AI-generated outputs, leading to provable machine learning and trustworthy AI. Our developed scheme computes AI predictions, cryptographic hashes of model inputs, and data hashes to immutably store them on a blockchain via smart contract (SC) using our novel input-output cryptographic hashing technique. This introduces a deterministic tokenization and canonical hashing pipeline that binds each GDSC2 drug–cell line input and its AI prediction output into a salted, on-chain verifiable commit. Later, a verification process has been committed by blockchain’s immutability and cross-checking via audit logs, which allows any stakeholder to independently confirm that a specific prediction originated from a known model and reliable source of data without exposing sensitive genomic content, ensuring both the verifiability and honesty of audits to serve the purpose for addressing AI post hoc tampering issue. The experimental results using genomic data inputs derived from the GDSCv2 dataset demonstrate the proposed model’s capability to train a Random Forest Regressor (RFR) for accurate AI-driven drug sensitivity prediction, achieving an R² of 0.979. Furthermore, 5-Fold Cross-Validation yielded a consistent mean R² of 0.977 ± 0.001, highlighting the model’s strong reliability, robustness, and generalization performance across multiple data partitions. Later, the model can store these predictions on-chain with due patient consent to verify or audit, detect tampering, ensure transparency, and verifiability up to 70% through an audit trial integrity test conducted for 10 samples from the dataset. The findings support the model’s applicability in high-stakes personalized medicine and biomedical environments where verifiable AI predictions are paramount.</p>