Abstract
This disclosure describes a method of governing AI agents that act continuously and without being prompted, such as agents running on a personal AI device or an organization's servers. An agent may take an unprompted action only under a mandate: a signed, machine-readable grant of authority from its owner that states what kinds of action are allowed, with which counterparties, within what budget, during what hours, and until when. Every unprompted action must cite the mandate it relies on, and the governance point that admits actions refuses any that cite no valid mandate or exceed one. Mandates are leased, not granted forever: each has a short term and lapses unless the owner renews it, and renewal requires the owner's verified presence at declared intervals. When a mandate lapses, the agent does not stop thinking or watching, but it may only propose actions for the owner to approve. Budgets, counts and counterparties are tracked cumulatively across all actions under a mandate. If the owner cannot be reached for a long period, mandates narrow step by step. Prompted actions, taken at the owner's direct request, are distinguished from unprompted ones and recorded as such. Every mandate, renewal, lapse, citation and refusal is recorded in a tamper-evident ledger, so the owner can always see what the agent did on its own and under what authority.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Regulator™: Authority Lease for Always-On AI Agents, with Signed, Expiring Mandates that Every Unprompted Action Must Cite", Technical Disclosure Commons, (October 13, 2026)
https://www.tdcommons.org/dpubs_series/12171