Abstract

This disclosure describes a method of governing payments made by AI agents. An agent's spending limits are bound to its certified credential and recorded in a public registry: limits per payment, per period and per class of payee or purchase, and lists of permitted and excluded payees. The limits are enforced at the point of payment by the payment service, which verifies the agent's credential and current limits before authorising each payment, and also by the agent's own runtime. Limits may be tightened at once by the agent's owner, its accountable SI Manager, the payment service or an SI Officer placing a hold, but may be loosened only with delayed two-party approval. A payment above a declared threshold, or to a payee not seen before, requires the signed approval of the agent's SI Manager within a set time, failing which it is declined. Every payment, approval, refusal and change of limit leaves a signed receipt in the registry, visible to the owner and to auditors. Spending stops automatically when the agent's credential is suspended, its SI Manager binding lapses, or the twin on which its spending behaviour was rehearsed lapses, apart from declared essential payments. A payee can verify, before accepting an order, that the agent is certified, that the order is within its limits and who answers for it. Disputed payments are resolved on the signed receipts.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Share

COinS