MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202611113242 A) filed by Moksh Kulshrestha; Dr. Priti Sharma; Ms. Ritu Malhotra; Dr. Gunjan Srivastava; Dilip Kumar; Ms. Amita Shukla; and Ms. Rakhi Singh on September 22, 2026, for System For Secure Artificial Intelligence Agent Task Execution Using Tool-Permission Scoring And Runtime Guardrails.
Inventors include Moksh Kulshrestha; Dr. Priti Sharma; Ms. Ritu Malhotra; Dr. Gunjan Srivastava; Dilip Kumar; Ms. Amita Shukla; and Ms. Rakhi Singh.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: A system for secure artificial intelligence agent task execution is disclosed. An agent-generated tool invocation is normalized with identity, task, provenance, resource and destination context and is evaluated against a machine-readable tool capability descriptor. A permission scoring engine computes an invocation-specific tool- permission score from security-relevant factors, and an execution-envelope generator converts the score and mandatory policy constraints into a parameter-level, machine-enforceable execution envelope. A controlled tool gateway binds the envelope to the invocation and dispatches the tool only within the authorized scope. Runtime guardrails monitor execution events, including resource access, destinations, process creation, data classification, side effects and cumulative task risk, and can continue, constrain, pause, terminate, revoke or roll back execution when runtime state departs from the authorized scope. Provenance labels limit authority propagation from untrusted retrieved content, while an integrity-protected ledger records decision evidence and execution outcome. The architecture thereby separates probabilistic agent planning from deterministic enforcement at the tool boundary. Accompanied Drawing [FIGS. 1-2].
Disclaimer: Curated by HT Syndication.
