MUMBAI, India, Aug. 24 -- Intellectual Property India has published a patent application (202641098018 A) filed by Keshav Memorial Institute Of Technology on August 13, 2026, for Red Team Web Framework: Security Testing And Assessment Framework.
Inventors include Mr. P Kiran Kumar; Mr. Narender Ravula; Ms. Arti Budhiraja; Ms. A. Gouthami; Mr. V Harivaths; Mr. Raghu Sattu; Mr. Mohammed Yasin Ali; and Mr. M Rishikesh.
The application for the patent was published on August 21, 2026, under issue no. 34/2026.
Abstract: Modern AI applications increasingly integrate Large Language Models (LLMs) as core components, introducing new security challenges such as prompt injection, jailbreaks, data exfiltration, and model manipulation. This project presents a web-based LLM Red Team Framework that automates adversarial testing and strengthens LLM defences through reproducible experiments. The framework enables security teams and developers to design, execute, and analyze red-team campaigns against both hosted and local LLM endpoints. It features a plugin-based architecture consisting of attack modules (e.g., prompt-injection, obfuscation, canary exfiltration) and defense modules (e.g., sanitizers, context filters, policy-check agents). An integrated experiment runner conducts deterministic and randomized tests, records complete audit logs, and generates quantitative security metrics such as Attack Success Rate and Leakage Score. Implemented as a modern, containerized web application, the system ensures safety and governance via sandboxed environments, role-based access control, and explicit target consent. Deliverables include a functional web UI, attack/defense plugin catalog, Dockized local testbed with an opensource model, and automated report generation. This framework transforms LLM adversarial testing into an operational defense workflow—allowing organizations to continuously measure, break, and harden their AI systems.
Disclaimer: Curated by HT Syndication.
