MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202641123766 A) filed by Vellore Institute Of Technology on October 15, 2026, for Model Integrity Authentication System For Machine Learning Inference Endpoints Via Behavioural Fingerprinting.
Inventors include Gayathri S; Rishi Chalana; Yash Kashyap; and Mithun Vijayan.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: ABSTRACT MODEL INTEGRITY AUTHENTICATION SYSTEM FOR MACHINE LEARNING INFERENCE ENDPOINTS VIA BEHAVIOURAL FINGERPRINTING A model integrity authentication system (100) includes one or more processors, a memory storing instructions, and a network interface through which a target model endpoint (126) is queried. A probe engine (110) generates a probe set comprising boundary probes, confidence probes, and latent-sensitivity probes for the target model endpoint (126). A fingerprint vault (112) stores a behavioural fingerprint (310) comprising a mean probability matrix (312), a top-2 rank matrix (314), and raw un-binned confidence values (316). A runtime verifier (114) periodically re-queries the target model endpoint (126) and computes a composite integrity score (418) based on Kullback-Leibler divergence, cosine similarity, maximum mean discrepancy, and top-2 rank agreement metrics. An alert engine and audit log (116) raises an alert when the composite integrity score (418) falls below a threshold.
Disclaimer: Curated by HT Syndication.
