MUMBAI, India, Nov. 2 -- Intellectual Property India has published a patent application (202641124747 A) filed by Vellore Institute Of Technology on October 19, 2026, for Divergence-Scoped Adaptive Re-Encryption For Model-Version-Driven Protection Of Nlp Detected Pii.
Inventors include Mohana Sundari L; Oorjaa Sonthalia; and Kaushik Sinha.
The application for the patent was published on October 30, 2026, under issue no. 44/2026.
Abstract: A computer-implemented system and method are disclosed for selectively maintaining cryptographic protection of Personally Identifiable Information (PII) within a text corpus following an update of a machine-learning-based classification model. The system determines a per-span divergence between tagging generated by an original model and re-tagging generated by a newer model based on boundary overlap, entity-type agreement, and confidence delta. A signed calibration mechanism cryptographically binds a re-encryption threshold to a specific ordered pair of model-version hashes representing a directional transition from a first model version to a second model version. The system further maintains a per-span lifecycle comprising ACTIVE, RETIRED, and PENDING_REVIEW states, wherein spans associated with retired classifications are subjected to mandatory review. A low-cost canary pre-flight employs anchored sampling and unanchored blind-spot probing to determine whether full-corpus re-tagging is warranted. Based on the evaluated divergence and validated calibration, cryptographic operations are selectively performed for affected spans, thereby reducing unnecessary corpus-wide re-encryption while maintaining updated PII protection.
Disclaimer: Curated by HT Syndication.
