Abstract
This disclosure describes a method of changing the AI model that controls a governed system, such as a personal AI device, an enterprise agent or a robot, so that a new model never takes control unchecked. A candidate model is admitted for evaluation only if its signed lineage record verifies and it comes from an authorized source. It then runs in shadow: it sees the same inputs as the model in control and proposes the actions it would take, but none of its proposals are executed. The governance point records where the candidate's proposals differ from the current model's, and where either would have been refused by policy. Safety-relevant differences are referred to an independent second opinion from a verifier of different lineage. The candidate must pass a declared test set of safety cases for the system's class. A plain-language summary of what changes is shown to the owner or responsible operator, whose consent is required. On consent, a restore point of the current model is preserved, and the candidate takes control at a raised governor level, stepping down to normal only as it builds a clean record. Declared triggers return control to the previous model automatically. Urgent security fixes follow a shortened but never skipped path. Every step is recorded in a tamper-evident ledger.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Regulator™: Model Change Control for AI Systems, with Shadow Operation, an Independent Second Opinion, Verified Lineage and Consent before a New Model Takes Control", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12174