MUMBAI, India, Oct. 26 -- Intellectual Property India has published a patent application (202644123315 A) filed by Sisa Information Security Pvt Ltd on October 14, 2026, for System And Method For Fortifying Generative Artificial Intelligence (ai) Models.
Inventors include Anjan Krishnamurthy; and Renju Varghese Jolly.
The application for the patent was published on October 23, 2026, under issue no. 43/2026.
Abstract: ABSTRACT A system and method for fortifying generative artificial intelligence (AI) models is disclosed. The system integrates automated scanning, adversarial testing, and selective fine-tuning within a continuous feedback loop. A model scanning subsystem performs static and dynamic analysis to detect vulnerabilities in the generative AI model and generate a vulnerability map. An adversarial assessment engine executes the generative AI model with exploit libraries and adversarial prompts to produce an exploit surface map identifying toxic responses. A model fine-tuning subsystem isolates and retrains toxic layers or activation pathways of the generative AI model using alignment datasets of adversarial prompts paired with benign responses. Embedded guardrails and iterative validation prevent behavioral regression and ensure the generative AI model alignment. The system thereby enables continuous security hardening, resilience against prompt injection, bias, and hallucination, and audit-ready assurance of trustworthy generative AI model deployment. FIG. 2
Disclaimer: Curated by HT Syndication.
