MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202611073941 A) filed by Satysaai Raghava Krishna Akkapeddi; and Sahil Tripathi on June 15, 2026, for Autonomous Self-Refuting Qa Architecture For Adversarially Robust Financial Threat Verification.
Inventors include Satysaai Raghava Krishna Akkapeddi; and Sahil Tripathi.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: The present invention relates to an antonomous self-refuting Question-Answering (QA) system configured according to Helpful, Harmless, and Honest (HHH) operational objectives for adversarially robust verification of financial threats. Conventional QA systems, including Retrieval-Augmented Generation (RAG) systems, transformer-based financial QA systems, and multimodal Large Language Models (LLMs) such as GPT-4 [13], produce responses from retrieved or generated content without an internal challenge proce dure and show susceptibility to prompt injection, retrieval corruption, and data-poisoning conditions within financial decision- making environments. The present invention addresses these limitations through a dual-stage self-refutation process in which an initial response is subjected to an automatically gencrated counter-query prior to issuance of a final HIIH-aligned output [3]. The present invention further comprises an adversarial input detection module trained using CICIDS2017 [14], a financial threat scenario generation component derived from the MITRE ATT&CK framework [2]. and a settlement validation module configured for ACH, RTP. and FedNow transaction workflows and evaluated using ConvFinQA [7] and FinanceBench [8]. The disclosed present invention provides a verification mechanism for adversarial thrcat assessment within financial infrastructure operating under high-consequence conditions.
Disclaimer: Curated by HT Syndication.
