MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202611101715 A) filed by Panipat Institute Of Engineering And Technology Piet on August 22, 2026, for System And Method For Secure Execution Of Large Language Model Agents Using Dynamic Tool Permission Control And Multi-Stage Action Verification.
Inventors include Ritika Sharma; Dr. Vaishali Mehta; and Piyush.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: [051] The present invention discloses a system and method for secure execution of Large Language Model agents using dynamic tool permission control and multi- stage action verification. The invention receives user execution requests through an agent interface, evaluates contextual information including user identity, operational intent, environmental conditions, and organizational policies, generates temporary execution permissions, verifies proposed tool actions through semantic validation, dependency analysis, parameter inspection, and policy compliance assessment, and executes authorized operations under continuous runtime monitoring. A policy enforcement module dynamically regulates execution according to configurable security rules while an audit intelligence module generates immutable execution records supporting compliance, governance, and forensic investigation. The invention significantly improves execution security, prevents unauthorized tool usage, reduces risks associated with prompt manipulation and adversarial inputs, enhances transparency, and enables reliable deployment of autonomous LLM agents across enterprise computing environments. Accompanied Drawing [FIGS. 1-2]
Disclaimer: Curated by HT Syndication.
