Abstract

Abstract:

The problem space AI agents are starting to move money without a person approving every step. They prepare purchases, pay invoices, buy machine services and send stablecoin transfers. At the same time, card networks, banks, blockchains and central banks in Europe, China and elsewhere are building faster and more programmable payment rails: instant credit transfers, tokenized deposits, stablecoins, the planned digital euro and the e-CNY. Today’s payment security layers answer five different questions:

  • Who is paying? Strong customer authentication and agent identity.
  • Is the credential valid? Tokens, cryptograms and signatures.
  • May this payment be made? Fraud scoring, spending limits and risk rules.
  • Did it process and settle? Networks, banks and blockchains.
  • What happened afterwards? Logs and records.

None of these, by itself, answers a sixth question: is this exact operation, with this exact payee, amount, currency and scope, the one that was authorized, after every service in the chain has handled it? Concrete failures this leaves open:

  • An agent is approved to pay Supplier A, but a compromised step swaps in a different account number, a larger amount or an extra fee.
  • A merchant-side service changes the capture conditions after the user’s approval.
  • A shared spending budget is copied into several machine agents, and each spends it (double spending).
  • A stablecoin transfer is signed for the wrong chain or address, and it cannot be recalled.
  • An instant transfer is submitted, the reply is lost, and a retry with a new authority pays twice.
  • A wallet or chat flow in a super-app reuses an approval from another conversation.
  • Several institutions take part in a cross-border or multi-leg settlement, and no one can show who authorized which leg.

These problems get worse with AI agents, because an instruction can be sent correctly yet scoped wrongly, and once money is signed or sent, a log cannot undo it. The usual objection is speed: card checkout, instant payments and micropayments have short deadlines, and a heavy check on every step would hurt.

The technical solution The disclosure proposes a narrow, separate check for consequential payment instructions only, in the style of a policy decision point and a policy enforcement point:

  1. A payment is first held as a pending operation. Nothing moves yet.
  2. An isolated validator rebuilds what the instruction will really do and checks it against the payer’s mandate, the beneficiary, amount, currency, scope, spending limits, policy and revocation.
  3. If it passes, a narrow, single-use, sender-constrained capability is issued, tied to that exact instruction and that exact enforcement point.
  4. The enforcement point (a bank payment-instruction interface, a merchant or orchestration backend, a signing or custody service, a smart-account policy module or a ledger-commit interface) verifies that capability before money moves. An edited, reused or out-of-date instruction is refused.
  5. A receipt states exactly which event it proves. It is evidence of authorization only, not of clearing, credit, blockchain confirmation or legal settlement.
  6. A timeout is treated as an unknown outcome. The system does not assume failure, does not issue a second authority, and retries only with the same idempotent authority, then reconciles.
  7. Machine micropayments use a protected budget with local checks and atomic consumption. Offline digital money uses bounded, expiring authority loaded earlier.
  8. Expensive preparation is done ahead (cold path). Only a few cheap checks run at the moment of effect (hot path). There is no new AI inference, no new payment rail and no change to blockchain consensus.
  9. Integration levels run from a gateway in front of a documented payment API, through hardware-protected signing and policy services, to bank or processor commit interfaces.

Scope of the disclosure Each product mapping states how the check would work (the operation, where the gate sits, what is checked, what is rejected, and what differs for that product):

  • Europe: Wero and the European Payments Initiative, SEPA Instant and TIPS, Adyen, Nexi, Worldline and Revolut, the ECB digital euro (online and offline), and the PSD3, PSR and MiCA reforms as publicly reported
  • China: Alipay (Ant Group), WeChat Pay, UnionPay, e-CNY (PBoC), mBridge, CIPS and CBETS, and Hong Kong FPS
  • Global: Visa (Intelligent Commerce, Trusted Agent Protocol), Mastercard (Agent Pay, Verifiable Intent, Multi-Token Network), Circle (USDC, EURC, Arc), Tether, Paxos, PayPal, Ripple, Stripe, Bitcoin and Lightning, Ethereum, Solana, Swift and BIS Project Agora

It includes flow diagrams, possible new technical effects, a critical-payments latency guide and real-world FAQs on checkout speed, instant payments, micropayments, offline money, validator outage, timeouts and unknown outcomes. It lists possible alignment with Anthropic, OpenAI, Google and Microsoft for AI agents that call payment tools.

Patent filing and disclosure record This disclosure is based on the Mothership application PCT/IB2026/055615, filed 4 June 2026 and publicly disclosed 16 July 2026. Its full text is in the TD Commons disclosure: https://www.tdcommons.org/dpubs_series/12542/ . It can also be found on PATENTSCOPE by searching “Sangam Das” or PCT/IB2026/055615 (pctib202655615). The application is pending and has not been granted.

AI-generated notice This document is AI-generated in substantial part and may contain mistakes. Please cross-check every fact, date, product detail, rule and filing identifier against the original filings and original citations before drawing any conclusion. It is an industry direction validation exercise that describes only how the invention could possibly align with the possible direction of the industries and companies named. It does not claim, state or imply that any company has used, copied, derived from or infringed the invention. It is not an endorsement or validation of any company, product, regulator or standard, and no affiliation is implied. The application is pending and not granted, and nothing here states the scope or validity of any claim. It is not legal, financial, investment or safety advice. The author holds a pending application, discloses that interest, and is open to an acquihire or strategic collaboration.

Keywords: AI safety; agent payments; exact-operation authorization; policy enforcement point; stablecoin; digital euro; e-CNY; PSD3; unknown outcome; offline digital money; worst-case latency

Licence: CC BY-NC 4.0

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Share

COinS