MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202611090819 A) filed by Noida Institute Of Engineering And Technology Niet on July 26, 2026, for System And Method For Adaptive Contamination-Resistant Benchmarking Of Artificial Intelligence Models.
Inventors include Ms. Ruchika; and Ms. Barkha Bhardwaj.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: A computer-implemented adaptive contamination-resistant benchmarking system (100) for artificial intelligence models comprises a dynamic item generation engine (110) synthesising evaluation items from parametric templates under expert-calibrated constraints; a cryptographic item vault (120) operating within a trusted execution environment (121) that stores live items in encrypted form and releases them solely to an ephemeral scoring channel (172); an expert calibration interface (130) aggregating reputation-weighted judgment from credentialed evaluators; a contamination detection unit (140) applying canary tokens, semantic fingerprints, and perplexity anomaly analysis; a signal validity analyzer (150) computing item response theoretic discrimination indices; a model evaluation orchestrator (170) mediating evaluated-model access through a secure application programming interface gateway (171); a procedural transparency ledger (180) recording tamper-evident governance metadata without disclosing item content; a
Disclaimer: Curated by HT Syndication.
