MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124756 A) filed by Vellore Institute Of Technology on October 19, 2026, for System And Method For Runtime Integrity Monitoring And Autonomous Quarantine Of Machine Learning Models Using Behavioral Representations And Causal Boundary Verification.
Inventors include Murugan Krishnamoorthy; and Sajal Varshney.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: The present invention provides an Autonomous AI Reliability Intelligence (AARI) framework for runtime integrity monitoring and autonomous fault containment of deployed machine-learning inference models. A model certification module (101) generates an integrity certificate (102) comprising model-specific behavioural representations, certified statistical parameters, causal-boundary constraints, and integrity thresholds. During inference, a per-inference behavioural verification module (103) generates a compressed behavioural representation and determines deviation from the certified behavioural boundary. A causal boundary verification module (104) evaluates runtime feature-sensitivity relationships against the certified causal constraints. A reliability-state determination module (105) evaluates behavioural and causal integrity measurements over configurable observation windows and determines a reliability state. Upon persistent or severe integrity violation, an autonomous model quarantine module (106) isolates the affected model, while an autonomous fallback routing module (107) redirects inference requests to a pre-validated fallback model. An integrity evidence recording module (108) records integrity measurements, state transitions, quarantine events, and recovery actions, providing autonomous model reliability and service continuity.
Disclaimer: Curated by HT Syndication.
