Intellectual Property India Publishes Patent Application for System And Method For Secure Hardware-Attested Verification Of Per-Shard Integrity And Path Trust During Machine Learning Model Loading.

MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202631105408 A) filed by Amit Kumar Gupta on September 01, 2026, for System And Method For Secure Hardware-Attested Verification Of Per-Shard Integrity And Path Trust During Machine Learning Model Loading..

Inventor includes Amit Kumar Gupta.

The application for the patent was published on September 11, 2026, under issue no. 37/2026.

Abstract: ABSTRACT System and Method for Secure Hardware-Attested Verification of Per-Shard Integrity and Path Trust during Machine Learning Model Loading. Technical summary: The invention relates to a system and method for verifying the integrity and provenance of individual shard files during the loading of large machine learning model artifacts. Conventional approaches verify model authenticity only at the level of the complete model package or repository, typically using a single cryptographic signature or hash covering the entire artifact set. This leaves a gap during the actual loading process, where a manifest or index file directs a loader to fetch and read individual shard files by name and file path. If the manifest or the file path resolution logic is tampered with, a loader can be induced to read an unintended or malicious file, even though the overall repository signature appears valid. The proposed invention introduces a verification step performed at the moment each individual shard is about to be loaded into memory, rather than only at the point of initial download or repository level validation. A hardware backed security module, such as a trusted platform module or secure enclave, holds a cryptographic key used to generate and verify a binding between each shard's content hash and its declared file path, as specified in the manifest. Before a shard is deserialized and loaded, the loader queries the hardware module to confirm that the shard's actual file path and content hash match the hardware attested expected values. If a mismatch is detected, loading of that specific shard is halted, preventing path manipulation or shard substitution attacks, while shards that pass verification proceed as normal, minimizing performance impact. The technical effect is an improvement in the security of the model loading process itself, specifically preventing unauthorized file access arising from manifest or path tampering, achieved through a defined hardware and software verification architecture rather than a purely algorithmic or software only check.

Disclaimer: Curated by HT Syndication.