MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202611090817 A) filed by Noida Institute Of Engineering And Technology Niet on July 26, 2026, for Adaptive Neuro-Symbolic Trust Mesh For Distributed Predictive Safety In Large Language Models.
Inventors include Mr. Ashish Kumar; and Mr. Amit Kumar Yadav.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: An adaptive neuro-symbolic trust mesh system (100) for distributed predictive safety in large language models comprises an oversight orchestrator (120) interposed between a user interface module (110) and a target language model (180). The system (100) decomposes safety assessment into four categorical gates: an intent analysis gate (130) recovering operational objectives, a zero-trust context verification gate (140) treating asserted authorities as unverified evidence, a consistency tracking gate (150) detecting multi-turn escalation patterns, and an output risk assessment gate (160) inspecting candidate responses. An escalation decision engine (170) combines gate verdicts through disjunctive escalation logic wherein any danger verdict blocks the interaction. Three pre-generation gates (130, 140, 150) block before the target model generates while the output gate (160) serves as a post-generation backstop. A conversation memory store (190) maintains dialogue history and an audit module (200) provides human-re
Disclaimer: Curated by HT Syndication.
