Abstract
Current Large Language Model (LLM) orchestration frameworks cannot cryptographically prove which end-client generated a request, trace, memory update, or provenance entry. This submission introduces a cryptographically enforced provenance and runtime isolation technique that binds every execution artifact to the originating client, preventing cross-tenant data leakage, cache contamination, audit ambiguity, and compliance failure, while enabling scalable shared backends.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kirti, Amisha and Alam, Nazrulhasan Nishat, "CLIENT-BOUND PROVENANCE AND SESSION ISOLATION FOR MULTI-TENANT LLM ORCHESTRATION FRAMEWORKS", Technical Disclosure Commons, (March 19, 2026)
https://www.tdcommons.org/dpubs_series/9570