MUMBAI, India, Feb. 13 -- Intellectual Property India has published a patent application (202541123873 A) filed by Ballari Institute Of Technology And Management, Ballari, Karnataka, on Dec. 9, 2025, for 'multi agent artificial intelligence system for autonomous detection and mitigation of adversarial or.'
Inventor(s) include Dr. Girish Kumar D; Dr. Yadavalli Basavaraj; Dr. Sadyojatha Kalapur Mutt; Dr. K S Shiva Kumar; and Dr. Pani Rama Prasad.
The application for the patent was published on Feb. 13, under issue no. 07/2026.
According to the abstract released by the Intellectual Property India: "The invention provides a multi-agent artificial intelligence framework for detecting and preventing jailbreak attacks in large language model (LLM) systems. A central processing unit (CPU) (001) receives user input and coordinates processing across several autonomous Al agents. A graphics processing unit (GPU) (002) generates text embeddings using a multilingual embedding model (002a) to support neural-network operations. An Input Classification Agent (003), running on the 10 GPU, uses a convolutional neural network to analyze embeddings and produce an initial assessment of possible adversarial or jailbreak intent. A Threat Detection Agent (004), executed on both CPU and GPU, applies an ensemble of a gradient-boosted decision tree and a support vector machine to generate a secondary jailbreak-likelihood evaluation. A Response Validation Agent (005) uses a rule-based safety engine and a fine-tuned BERT model to determine whether 15 the LLM's output is safe. A Reinforcement Learning-Feedback Agent (006), stored in memory (007), continuously updates system parameters to improve detection accuracy."
Disclaimer: Curated by HT Syndication.
