In Depth: When AI Spends, Card Networks Scramble to Keep Payments Safe

1. AI agents are evolving from question-answering tools to autonomous shoppers handling transactions like ordering bubble tea, booking flights, and buying insurance, surging transaction volumes and posing payment security challenges for finance [para. 1][para. 2].

2. China UnionPay notes that extending payments to AI agents reshapes payment logic; to counter rogue AIs draining accounts, networks build new infrastructure [para. 3][para. 4].

3. UnionPay launched Agentic Payment Open Protocol (APOP) in Shanghai in early April, verifying transactions like Umetrip flights and Zhipu AI coffee orders, with 19 partners including banks and AI developers; it verifies AI identity, user intent, and secures payments [para. 4][para. 5].

4. Mastercard rolled out certified AI agent payments in March across Australia, Singapore, Latin America, and Hong Kong, building trust layers for compliant payments without breaching AML/fraud rules [para. 6].

5. AI payments break the "real human assumption," requiring overhauls in risk controls and liability; card networks face trust deficits in user authorization and AI authentication [para. 7][para. 8].

6. UnionPay outlines three phases: current assisted transactions; conditional autonomy (e.g., phone recharges); full autonomy interpreting vague requests, needing risk/refund updates [para. 9][para. 10].

7. Unlike integrated giants like Alibaba, card networks use external AI devs, complicating liability in chains like UnionPay's Hong Kong eLife pilot with CardInfoLink and BoC Pay [para. 12][para. 13].

8. UnionPay proposes "Know Your Agent" (KYA) registration platform like ATM certifications, plus KYM, operator verification, and user real-name auth for traceability and security [para. 14][para. 15][para. 16].

9. Mastercard requires agent authentication for tokens and merchant API certification [para. 17]; tokenization with biometrics/limits prevents abuse, with human verification for high-value deals [para. 19][para. 21][para. 22][para. 23].

10. Anti-fraud adapts with digital signatures, contextual anomaly detection; privacy risks from malicious agents noted by Yang Tao; open protocols like APOP enable interoperability [para. 25][para. 26][para. 27][para. 28].

11. Mastercard predicts rapid AI commerce growth, multi-agent systems, needing standards; partners like OpenAI/Anthropic adhere to rules [para. 29][para. 30].

12. UnionPay's "new four-party model" adds wallets/AI agents to cardholders/merchants/issuers/acquirers, paving for unified global standards and adoption [para. 31][para. 32][para. 33].