Intellectual Property India Publishes Patent Application for Blockchain-Based Ai Model Ownership Verification Using Behavioral And Activation Fingerprinting

MUMBAI, India, Oct. 5 -- Intellectual Property India has published a patent application (202641113832 A) filed by Siva Shankari. M; K. Ashokkumar; Shraddha Satish Kashid; R. Pradeepa; Dr. K. Saranya; and Dr. B. Gunasundari on September 23, 2026, for Blockchain-Based Ai Model Ownership Verification Using Behavioral And Activation Fingerprinting.

Inventors include Siva Shankari. M; K. Ashokkumar; Shraddha Satish Kashid; R. Pradeepa; Dr. K. Saranya; and Dr. B. Gunasundari.

The application for the patent was published on October 02, 2026, under issue no. 40/2026.

Abstract: Blockchain-Based AT Model Ownership Verification System for establishing and verifying the ownership and provenance of deep learning models uses behavioral and activation fingerprinting. A trained ResNet-50 convolutional neural network (CNN) is used as the reference AT model. A predefined probe dataset is applied to the model to generate a behavioral fingerprint consisting of prediction probabilities, class responses, confidence scores, and prediction consistency. For activation fingerprinting, intermediate feature maps are extracted from selected ResNet-50 layers using Grad-CAM and Global Average Pooling (GAP). The resulting activation vectors are reduced using Principal Component Analysis (PCA) and converted into a compact fingerprint. The behavioral and activation fingerprints are combined and processed using SHA-256 cryptographic hashing to generate a unique model fingerprint. The fingerprint, model identifier, owner information, timestamp, and provenance metadata are registered on a Hyperledger Fabric permissioned blockchain through a Solidity/Chaincodebased smart contract. During verification, the submitted model is evaluated using the same probe dataset, and its newly generated fingerprint is compared with the registered fingerprint using cosine similarity, Euclidean distance, and Hamming distance. A predefined verification threshold determines fingerprint correspondence. The proposed system provides a tamperresistant, traceable, and cryptographically verifiable mechanism for Al model ownership, provenance tracking, and detection of unauthorized model copies or modified versions.

Disclaimer: Curated by HT Syndication.