Abstract

Autonomous AI agents, including large language model (LLM) agents, increasingly originate payment authorization requests under stored, cryptographically signed mandates derived from natural-language instructions. Existing frameworks verify machine-enforceable constraints — amount, payee, merchant — at transaction time; existing art separately teaches similarity-based transaction screening, single-use transaction-linked authorization credentials, decoupled out-of-band authentication, versioned credential supersession with freshness proofs, and attested display of transaction details. No reference teaches the lifecycle coupling specified here: (1) suspending a payment authorization request when a similarity score between a salted, committed embedding of the natural-language instruction — bound into the signed mandate with a pinned scoring stack and threshold — and an embedding computed over the canonical transaction commitment fails the mandate's threshold, producing a signed divergence record rather than a decline; (2) a MandatePatch, a human-signed object modifying payment-authorization scope for one transaction only, identifying the violated predicates and exact checkout, consumed on use, while leaving the base mandate unmodified for sibling and future transactions; (3) superseding a stored mandate through cryptographic predecessor lineage with deterministic resolution of in-flight transactions at amendment or revocation across heterogeneous payment rails; (4) resuming a suspended agent execution state only upon expiring out-of-band confirmation whose signed response simultaneously authorizes the delta and binds to that exact state; and (5) presentation-state attestation binding a canonicalized hash of the rendered approval interface into mandate and patch artifacts. A zero-knowledge enforcement backend is specified, with a minimal reference circuit measured at 6,591 R1CS constraints.

Creative Commons License

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

circuit-reference-v2.2-TDCOMMONS.tar.gz (1412 kB)
Reference ZK circuit (Circom/Groth16): sources, compiled R1CS, and sample proof supporting the measured constraint counts in Section 7 and Appendix A. Self-verifying; see NOTICE.md for provenance.

Share

COinS