MUMBAI, India, Aug. 12 -- Intellectual Property India has published a patent application (202621036767 A) filed by Persistent Systems on March 26, 2026, for System And Method For Agent-Based Iterative Obfuscation With Hybrid Local-Cloud Inference For Privacy-Preserving Language Models.
Inventor includes Mr. Nitish Shrivastava.
The application for the patent was published on August 07, 2026, under issue no. 32/2026.
Abstract: A system and method for privacy-preserving processing of sensitive organizational data during interactions with large language models are disclosed. The system comprises a gateway proxy (110) configured to intercept requests from enterprise applications (101) prior to transmission to external LLM (102). A agent orchestration engine (120) including a detection agent (121), semantic analysis agent (122), replacement generation agent (123), and quality assurance agent (124) identifies sensitive tokens, evaluates contextual sensitivity, and generates semantically equivalent replacements using a local SLM (130). A sensitive token cache (140) and a placeholder mapping registry (150) securely store encrypted mappings between original and replacement tokens. A response reconciliation engine (160), including placeholder swap module (161), gap detection module (162), semantic gap fill module (163), and coherence verification module (164), reconstructs original information in responses. An iterative refinement mechanism ensures transformation quality while preserving semantic integrity and preventing sensitive data exposure.
Disclaimer: Curated by HT Syndication.
