Abstract
This disclosure describes a governed method for ending the service life of certified AI systems, AI agents and robots. Retirement begins with a notice recorded in a public registry, stating the date of retirement, the reason, the successor (if any) and the plan, given a declared minimum period in advance except where retirement is forced by a safety finding. The plan is rehearsed on the system's certified twin before it is carried out. People, systems and other agents that depend on the retiring system are identified from the registry and are handed over to a successor or notified, so that no dependant is left without service unannounced. Custody of the system's governed data passes to a named successor custodian and is proven by a drilled restore. The system's credentials are revoked at the moment of retirement, and the revocation propagates to every service that relies on them, so that the retired system, or anything impersonating it, can no longer act. Recorders, signed records and audit evidence are retained for a declared period under independent custody. A retired robot is cleared of data, its batteries and hazardous parts are removed through certified channels, and its components are recorded as leaving service in the supply-chain records, or as refurbished under a new record that keeps its history. The registry record of the system is closed, showing its final status, but is never erased.
Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Sunset™: Governed End of Life for AI Systems, Agents and Robots, in which Retirement Is Announced and Rehearsed in Advance, Dependants Are Handed Over before Service Ends, Data Custody Passes to a Named Successor and Is Proven by a Restore, Credentials Are Revoked Everywhere at Once, Recorders and Records Are Retained for a Declared Period, Physical Machines Are Disposed of through Certified Channels, and the Registry Record Is Closed but Never Erased", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/13003