The core insight of the research is that existing approaches tend to treat these technologies separately. Blockchain frameworks excel at providing decentralized trust, transparency, and tamper-resistant record-keeping, but they lack the intelligence to detect threats in real time. Machine learning models can identify anomalies and classify risks with impressive accuracy, but they typically operate on centralized data stores that are themselves susceptible to manipulation and single points of failure. Encryption schemes like AES and RSA safeguard data in transit and at rest, but most implementations do not integrate decentralized verification or intelligent threat detection. The result, according to Al-Abassy, is a fragmented security landscape in which each technology addresses only part of the problem while leaving critical gaps in confidentiality, integrity, privacy, and trust.

To close these gaps, the proposed framework deploys a lightweight Random Forest classifier as its first line of defense. Before any file is encrypted or registered on the blockchain, the classifier examines file-level features such as size, type, and extension, assigning each file to one of three risk categories: Low, Medium, or High. This classification is not merely academic. It determines the level of protection each file receives, allowing the system to apply adaptive security rather than a one-size-fits-all approach. Common document formats and non-sensitive files are flagged as Low risk, executable formats and moderately sensitive content as Medium risk, and encrypted archives, highly sensitive medical records, or files with multiple security-sensitive properties as High risk. The classifier was trained on an internally generated dataset of 10,000 file samples, stratified into 4,000 Low-risk, 3,000 Medium-risk, and 3,000 High-risk categories, with 80 percent used for training and 20 percent reserved for independent testing.

Once a file has been classified, the framework applies AES–RSA hybrid encryption to secure it. AES, the Advanced Encryption Standard, is a symmetric cipher prized for its speed and efficiency in encrypting large volumes of data. RSA, named after its inventors Rivest, Shamir, and Adleman, is an asymmetric algorithm that excels at secure key exchange and authentication. In the hybrid scheme, AES handles the bulk encryption of the file content, while RSA encrypts and distributes the AES session keys, ensuring that even if an attacker intercepts the communication, the encrypted content cannot be retrieved without the proper decryption key. This division of labor is what gives hybrid encryption its power: AES provides high-speed protection for the data itself, while RSA guarantees that the keys protecting that data are exchanged securely.

After encryption, the framework registers the file on a lightweight private blockchain. Only essential metadata is stored on the ledger—file identifiers, timestamps, access details, and SHA-512 hash values—while the encrypted files themselves reside in cloud storage. Each block on the chain includes the timestamp, transaction details, current hash, and previous hash, creating a chronological, tamper-resistant record of every file’s lifecycle. A permissioned validation mechanism limits participation in transaction verification to approved entities, reducing the computational overhead that plagues public blockchains while maintaining the decentralized integrity verification, traceability, and immutable auditing that blockchain provides. The result is a system in which no single party can unilaterally alter the record of what was shared, when, and by whom.

The experimental results are striking. Across heterogeneous datasets containing files of different sizes and types, the framework achieved an average encryption time of 1.68 milliseconds, an average decryption time of 0.63 milliseconds, and an average blockchain verification time of 2.75 milliseconds. The Random Forest classifier reached 99 percent accuracy in multi-class risk classification, with precision, recall, and F1-score all averaging 99 percent across the three risk categories. Blockchain-based integrity verification achieved 100 percent accuracy. Every single security operation—whether encryption, decryption, or blockchain verification—completed in under 0.01 seconds, a threshold the researcher identifies as essential for real-time smart environment applications. Statistical analysis across thirty independent experimental runs confirmed the consistency of these results, with p-values below 0.001 for all measured metrics.

File-level analysis revealed that processing time scales with file size rather than file type. A small 120-kilobyte CSV dataset required just 1.21 milliseconds to encrypt, while a 5,400-kilobyte executable file took 2.03 milliseconds. A 6,800-kilobyte binary firmware file, classified as High risk due to its potentially dangerous format, required 2.45 milliseconds. This direct proportional relationship between file size and processing time—expressed mathematically as T ∝ S—demonstrates that the encryption and decryption algorithms handle data regardless of format, and that the system remains stable and effective across the full spectrum of digital content found in smart environments.

Ablation studies, in which individual components were systematically removed while the rest of the system remained intact, confirmed that the framework’s strength lies in the complementary contributions of its parts. Removing the AI classification module eliminated adaptive risk assessment. Replacing AES–RSA hybrid encryption with a baseline technique reduced the level of cryptographic protection. Removing the blockchain layer eliminated immutable auditability and traceability. Switching from SHA-512 to SHA-256 hashing modified the integrity verification method. In every case, the full configuration offered the most complete set of security, integrity, traceability, and intelligent risk-assessment features, demonstrating that the framework relies not on any single technique but on the synergy of all four.

Comparative analysis against existing security frameworks highlighted the balanced trade-off the proposed system achieves. Traditional IoT security frameworks relying solely on AES encryption reached only 65 percent detection accuracy because they lack intelligent detection capabilities. Machine learning-based intrusion detection systems improved accuracy to 78 percent but offered no encryption or decentralized trust layer. Blockchain-based storage security solutions provided strong data integrity but suffered from high computational overhead, with encryption times reaching 3.50 milliseconds. Deep learning security models achieved 85 percent accuracy but lacked blockchain-based integrity protection. The proposed framework, by contrast, combines 99 percent detection accuracy with millisecond-scale encryption, decryption, and verification times, all while maintaining full file and data integrity through SHA-512 hashing.

The implications extend beyond the laboratory. As smart cities, intelligent healthcare systems, and industrial IoT networks continue to expand, the need for security architectures that can guarantee confidentiality, integrity, privacy, and trust simultaneously becomes ever more urgent. The framework’s computational complexity analysis shows that each component scales efficiently: AI-based classification operates at O(n × f), AES encryption and decryption at O(m), SHA-512 hashing at O(m), and blockchain verification at O(b), where n represents the number of processed files, f the extracted features, m the file size, and b the blockchain verification operations. Al-Abassy suggests that future work will explore lightweight blockchain consensus mechanisms, edge intelligence, federated learning, privacy-preserving techniques, and post-quantum cryptography to further enhance scalability, resilience, and real-time secure data sharing. For now, the research offers a compelling demonstration that when artificial intelligence, hybrid cryptography, and blockchain are woven together rather than deployed in isolation, the result is a security architecture that is simultaneously intelligent, scalable, and trustworthy.

Subject of Research: An AI-driven blockchain framework with AES–RSA hybrid encryption for secure data sharing in IoT smart environments

Article Title: An AI-based blockchain framework with AES–RSA hybrid encryption for secure data sharing in smart environments

Article References: AbdulKareem Al-Abassy, S. (2026). An AI-based blockchain framework with AES–RSA hybrid encryption for secure data sharing in smart environments. Frontiers in Blockchain, 9, Article 1935632. https://doi.org/10.3389/fbloc.2026.1935632

Image Credits: AI Generated

DOI: 10.3389/fbloc.2026.1935632

Keywords: artificial intelligence, blockchain, AES, RSA, hybrid encryption, Internet of Things, smart environments, data security, anomaly detection, cybersecurity, AI-based, framework