The cardholder is now a machine: Mastercard and Visa wire AI agent identity into the global payment rail.

On the same day, Visa disclosed a $70 billion annualized stablecoin settlement run rate at Visa Payments Forum 2026 and announced a partnership with OpenAI for agent-initiated payments. Both card networks crossed the same threshold on the same day.

The June 10 announcements from Mastercard and Visa contain five discrete moves. Only the first is genuinely new.

Four of the five moves extend existing infrastructure. The credential is the new thing.

AP4M operates as three sequential layers: a human principal sets permissions; those permissions are recorded on-chain as an agent credential; the agent presents the credential to Mastercard's network for transaction execution and settlement.

Visa's parallel architecture uses an Agentic Directory (verified agent + merchant registry) and Agent Score (merchant readiness rating) instead of on-chain permission records, with OpenAI agents as the first live identity integration.

The addressable merchant base is 100 million. Card rail stablecoin settlement, as covered in the Mastercard and Visa briefings of June 4 and May 16, gave stablecoins access to the card network's settlement layer. AP4M extends a different thing: it gives AI agents access to every merchant who already accepts Mastercard, without those merchants integrating any new technology. The agent presents a Mastercard credential; the merchant's existing point-of-sale infrastructure handles the rest. The gap between "stablecoins can settle on card rails" and "AI agents can spend at 100 million merchants" is the gap AP4M closes.

The on-chain permission record is a compliance architecture, not a marketing feature. Recording spending rules on Polygon, Solana, or Base creates an auditable, portable, and independently verifiable authorization log. An enterprise deploying AP4M-credentialed agents can point to an immutable on-chain record of what each agent was authorized to spend, when, and with whom. For regulated financial services, procurement, and treasury functions, this is a meaningful audit trail that off-chain authorization systems cannot replicate. The stablecoin settlement leg is a convenience; the on-chain audit trail is the product for institutional adopters.

Visa's $70 billion figure reframes what card-network stablecoin settlement actually looks like. The April 29 investor disclosure reported $7 billion annualized for the nine-blockchain pilot. The June 10 figure of $70 billion covers the full Visa network. The disclosure gap confirms that card-network stablecoin settlement at meaningful scale was already running; the question is no longer whether the infrastructure works, but who is authorized to use it and for what transactions.

AP4M and Visa's agentic stack are launch announcements. The structural questions are not answered.

• No AP4M commercial volume disclosed. Thirty-plus partners have signed up. No transaction count, no settlement volume, no contribution to Mastercard's revenue has been published. The launch defines a standard; whether the market adopts it is unknown.
• GENIUS Act classification of agent-initiated transactions is unresolved. The OCC's June 12 draft PS-01 and PS-02 forms (covered in the June 15 briefing) address PPSI reserve and financial reporting. They do not address who is the "customer" when an AI agent, not a natural person, initiates a stablecoin transaction. Whether the agent's human principal, Mastercard, or the credential-issuing partner bears PPSI compliance responsibility is not settled.
• BVNK acquisition has not yet closed. Mastercard announced the $1.8 billion acquisition on March 17, 2026, pending regulatory approvals, with closing expected in late 2026. BVNK's $30 billion in annual stablecoin transactions across 130-plus countries is the infrastructure AP4M needs to reach emerging market merchants. Until the deal closes, AP4M runs on Mastercard's existing MTN settlement infrastructure.
• Visa and Mastercard agent credentials do not interoperate. An AI agent credentialed by Mastercard cannot transact on Visa's network and vice versa. Both card networks are building parallel, non-interoperable agent identity stacks. The duopoly pattern at the credential layer mirrors the historical duopoly at the card acceptance layer.

Three parallel build-outs are converging on programmable money from different sides. The wholesale side: BIS Project Agorá (May 27) proved atomic cross-border settlement using tokenized central bank reserves and tokenized commercial bank deposits. DTCC's Canton tokenization goes live in October 2026 for institutional equities and Treasuries. Fnality and Partior are targeting the bank-to-bank interbank leg. On June 5, eighteen US banks committed to a shared tokenized deposit network via The Clearing House (TCH) targeting H1 2027. All three are solving the same problem for the top tier of the financial system.

AP4M and Visa's agentic stack address a different part of the stack. DTCC Canton and TCH's tokenized deposit network are built for institutional actors settling large-value transactions in regulated financial markets. AP4M is built for AI agents making high-frequency, low-value payments to any of the 100 million merchants already accepting Mastercard. The two stacks are complementary, not competing. An agentic treasury management system might use DTCC Canton for the interbank leg and AP4M for vendor payments below a threshold, with the same underlying stablecoin.

The OCC's GENIUS Act final rule, due July 18, is the shared binding constraint. The rule determines which stablecoins qualify as Permitted Payment Stablecoin Issuers. USDC (Circle, conditional OCC charter) and RLUSD (Ripple, conditional OCC charter) are both AP4M settlement stablecoins. Their final GENIUS Act status depends on the July 18 rule. If either charter is not finalized by then, the stablecoin's eligibility in regulated card settlement is in question.

Mastercard and Visa spent eighteen months adding stablecoin settlement to the card rail. June 10 is the day they issued that rail to a new class of counterparty: AI agents with programmatic identity, on-chain spending rules, and access to over 100 million card-accepting merchants. The question is not whether agentic commerce will use card rails. It will. The question is whether the OCC's GENIUS Act final rule on July 18 places agent-initiated stablecoin transactions inside the regulatory perimeter, and whether Mastercard's BVNK acquisition closes in time for AP4M to reach the emerging market corridors where the economics of machine-speed micropayments are most compelling.

Watch three things:

• OCC GENIUS Act final rule, July 18. If the rule finalizes Circle's and Ripple's charters, USDC and RLUSD become unambiguously eligible for AP4M settlement. If the rule is delayed or the charters are not finalized, the two primary AP4M stablecoins sit in a compliance gray zone for any regulated issuer using them in agent transactions.
• AP4M volume at 90 days. The first public transaction count or settlement volume from any AP4M partner will reveal whether machine-speed micropayments can run on card rails without batch-window friction. The card economics of sub-dollar transactions at interchange fees have never been viable. AP4M's stablecoin leg sidesteps interchange; whether that makes micropayments economically viable at scale is the test.
• BVNK acquisition close and integration architecture. BVNK's $30 billion in annual stablecoin transactions and 130-country reach is what AP4M needs for the markets where programmatic micropayments matter most: cross-border gig economy payments, agent-to-agent settle in sub-Saharan Africa and Southeast Asia, automated FX conversion at the agent layer. The integration design will define whether AP4M is a developed-market product or a global one.