MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202621056913 A) filed by Sundeep Anand on May 05, 2026, for Deterministic Replayable Runtime System And Method For Artificial Intelligence Workload Execution With Transparent Model Interception, Tamper-Evident Audit, And Edge-Native Process Isolation.
Inventor includes Sundeep Anand.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: A runtime system and method for executing artificial intelligence inference workloads with deterministic replay, tamper-evident audit, and hardware-agnostic process isolation. A daemon process binds a unique Unix domain socket for each job run and starts a model interception server through which all model inference calls made by a job subprocess are intercepted, recorded, and served through a composable resolver pipeline comprising fixture cache, live model, and replay resolver stages. On freeze-mode replay, recorded responses are served from the original execution trace indexed by a monotonic call counter, exactly reproducing any prior run without invoking any model API. An audit log using BLAKE3 hash-chaining with model provider response identifier anchors provides cryptographically verifiable records of all model interactions. Subprocesses are isolated via Linux namespaces, seccomp-BPF syscall filters, and cgroups v2 limits without hardware virtualisation, enabling deployment on ARM64 edge hardware. A SHA-256 content- addressable fixture cache enables offline-first operation.
Disclaimer: Curated by HT Syndication.
