PayPal & Google Say Crypto Rails Are the Only Option for AI Agents

The clearest signal yet that the payments industry has accepted crypto as foundational infrastructure arrived not from a DeFi protocol or a blockchain startup — it arrived from a stage in Miami on May 10, 2026, courtesy of two of the most powerful names in mainstream technology. PayPal and Google Cloud told an audience at Consensus Miami that autonomous AI agents are constitutionally incapable of using the existing banking system, and that stablecoins, open payment protocols, and cryptographic multi-party custody are not merely interesting alternatives — they are the only realistic architecture for the coming wave of machine-to-machine commerce. The implications cascade far beyond crypto: they touch every merchant, every payment processor, and every regulator deciding right now what rules govern programmable money.

An Impossible Account Opening — and the Protocol Built to Replace It

The sharpest line of the entire Consensus panel came from Richard Widmann, Google Cloud's global head of Web3 strategy. Addressing why agentic commerce cannot simply bolt onto existing financial plumbing, Widmann was categorical: "An agent cannot get a bank account. It's not hard, it just is impossible," he said, citing both technological and regulatory barriers. The observation sounds almost comically simple, but it encapsulates a structural problem that no amount of API wrappers can fix. Know-your-customer rules require a legal person or registered entity; AI agents are neither. Card-network authorization flows require a human cardholder to authenticate; autonomous software operates without a human in the loop by design.

The consequence, Widmann argued, is that crypto is "a fantastic machine-readable interface for payments" — not because it is ideologically pure, but because it is the only payment layer an agent can interact with programmatically, without requiring human sign-in. Google's official announcement on its Cloud blog elaborates on the mechanism: AP2 uses cryptographically signed mandates — tamper-proof digital contracts that encode verifiable proof of user intent before an agent ever initiates a transaction. These mandates are bound to Verifiable Digital Credentials (VDCs), creating a non-repudiable audit trail. There are two types in the initial spec: a Cart Mandate, signed by the user at the moment of authorization and bound to a specific purchase, and a Payment Mandate, shared with the payment network and issuer to surface the agentic nature of the transaction so risk models can adapt.

Google has donated AP2 to the FIDO Foundation — the same standards body that manages passkey and passwordless authentication — and has recruited more than 120 partners to the protocol, including PayPal. Widmann explicitly compared the move to Coinbase's decision to give the x402 internet-native payment standard to the Linux Foundation, signaling that neither company believes locking agentic payment standards behind proprietary walls will create the trust enterprises need. "Open dialogues and open standards are really the foundation you need to build on," Widmann said.

On custody, Widmann described an architecture in which an agent holds only one shard of a two-of-three or three-of-five multisig key set, rather than a full private key. Google has extended its Cloud KMS platform to cryptocurrency custody to support this model, so the key management infrastructure enterprises already use for secrets and certificates can double as the custody layer for agentic wallets. The design principle is explicit: an agent "cannot simply unilaterally move funds or take action." That constraint is not just an engineering preference — it is the answer to every compliance officer's first question about autonomous spend.

This architecture dovetails directly with what we have covered in depth on this site. Our developer primer on how AI agents pay with x402 and EIP-3009 shows how programmable authorizations and spending limits can be baked into the wallet layer itself — AP2's mandate model is the enterprise-grade, compliance-aware counterpart to those lower-level primitives.

The 95/20 Chasm: Why Merchants Are the Weakest Link in Agentic Commerce

May Zabaneh, PayPal's Senior Vice President and General Manager of Crypto, brought the merchant side of the equation into sharp relief. PayPal runs an internal survey tracking AI agent traffic across its merchant network — a network that spans hundreds of millions of active accounts and millions of business integrations worldwide. The results are striking: 95% of merchants already see AI agent traffic on their sites, but only 20% have machine-readable catalogs that an agent can actually parse and act on. That 75-percentage-point gap is the near-term bottleneck for agentic commerce at scale, and it has nothing to do with crypto or protocol design — it is a merchant readiness problem that mirrors a challenge the industry has solved before.

Zabaneh's framing was deliberate and historically grounded. She positioned the agentic transition as the third commercial channel shift PayPal has navigated: from brick-and-mortar offline commerce, to web-based online stores, to mobile. Each transition required merchants to rebuild how they presented their inventory and accepted payment — and the merchants who moved early captured disproportionate share. "Merchants need to be ready for this next era," she said. The implication is that structured data — machine-readable product catalogs, agent-accessible APIs, schema-compliant inventory feeds — is now a competitive necessity, not a nice-to-have. PayPal's official agentic commerce page already shows a product called Store Sync, which surfaces merchant catalogs directly into LLMs including Microsoft Copilot and Perplexity — a direct answer to the readiness gap.

On the stablecoin side, Zabaneh was equally direct: PYUSD is "a very natural programmable layer for payments," particularly as commerce trends toward globalization, AI-native experiences, and tokenized assets. The positioning is not accidental. In March 2026, PayPal expanded PYUSD to 70 markets worldwide, making the dollar-backed stablecoin available across Asia-Pacific, Europe, Latin America, and North America. The global footprint matters enormously for agentic commerce: an AI agent shopping across borders on a user's behalf needs a single settlement currency that works in all jurisdictions without FX conversion friction.

The macro stablecoin context adds urgency. McKinsey estimates real stablecoin payments are running at approximately $390 billion annually, with B2B flows accounting for the majority — but the ECB has estimated that only roughly 0.5% of stablecoin volume represents organic retail-sized transfers, and roughly 6% of transactions go toward goods and services. The implication is that agentic retail commerce — the thing PayPal and Google are building for — is still a rounding error relative to total stablecoin volume. The standard-setting happening at events like Consensus Miami is precisely what bridges that gap from experiment to enterprise norm. For deeper context on this dynamic, our analysis of stablecoins' second wave powered by AI agents charts how the catalyst has shifted from DeFi speculation to machine-executed commerce.

Meanwhile, a parallel coalition has been assembling competitive infrastructure. As we reported earlier this month, AWS, Coinbase and Stripe's rival stablecoin payment rails — built around USDC, AgentCore, and x402 — represent a distinct but increasingly complementary approach. Where AP2 targets the enterprise merchant-to-consumer flow, x402 handles API-level micropayments between software services. The two standards are likely to coexist, much as TCP/IP and HTTP coexist within a common internet stack.

Liability Gaps, Regulatory Timing, and the Infrastructure Power Grab

For all the technical elegance of AP2 mandates and multi-party custody, both Widmann and Zabaneh acknowledged that the hardest problems in agentic commerce are not engineering problems — they are governance and liability problems. When asked directly about accountability, Zabaneh said the question of who is responsible if an agent makes a bad purchase is "definitely something that we have to think through as an industry." That candor should be read as a warning signal by any enterprise considering early deployment. The absence of settled legal doctrine around agentic liability means early movers face asymmetric downside: if an agent over-spends, double-orders, or is manipulated by a prompt-injection attack into an unauthorized transaction, the legal recourse framework simply does not exist yet.

Widmann's own sleepless concern pointed in the same direction from an infrastructure angle: the open question of "how do you onboard agents into all of the existing capital markets and infrastructure plumbing that powers payments and trading today" remains unsolved. Legacy clearinghouses, correspondent banking networks, and card scheme rulebooks were all written with human principals in mind. Retrofitting them for software principals is a multi-year institutional negotiation, not a sprint.

Regulatory timing creates both a risk and an opportunity. The Senate CLARITY Act markup set for May 14 could establish the first federal framework explicitly governing stablecoin issuers — a prerequisite for enterprise treasury teams to hold PYUSD or USDC on balance sheet without a laborious legal opinion process. If that legislation passes, it removes one of the most cited blockers to large-scale agentic commerce adoption. If it stalls again as it did in January, the runway extends by months or quarters, giving more time for technically capable but regulatory-risk-averse institutions to sit on the sidelines.

Meanwhile, the infrastructure layer is being locked in regardless of legislative outcome. Kraken's Payward entity is pursuing an OCC national trust charter; traditional banks are launching proprietary stablecoins at an accelerating pace. The entities that control stablecoin issuance, custody infrastructure, and protocol governance today will exert outsized influence over agentic commerce economics for years. That is the competitive logic driving both Google Cloud's decision to steward AP2 through FIDO and PayPal's decision to expand PYUSD to 70 markets before the regulatory framework is even settled. As we detailed in our analysis of Coinbase's pivot toward agentic infrastructure, every major platform with distribution scale is making the same bet: that whoever owns the payment rails for agents will capture more value than whoever built the agents themselves.

PayPal's official press release on its 70-market PYUSD expansion makes the commercial logic explicit: merchants using PYUSD can access payment proceeds within minutes rather than waiting days, improving working capital for cross-border commerce. That settlement speed advantage is entirely independent of the agentic layer — but when you combine instant settlement with programmable mandates and machine-readable catalogs, the compounding effect is what makes the architecture genuinely disruptive to Visa, Mastercard, and the correspondent banking corridors they depend on.

The Consensus Miami session was not a product launch or a funding announcement — it was something arguably more consequential: two of the world's largest technology and payments companies publicly committing their infrastructure roadmaps to the thesis that crypto rails are structurally necessary, not merely fashionable, for the next era of commerce. The observable signals to track from here are specific. First, watch AP2 partner announcements: if Visa, Mastercard, or JPMorgan join the FIDO consortium, the protocol transitions from a tech-company preference to an industry standard with real clearing-network teeth. Second, watch PYUSD supply growth on Solana and Ethereum L2s — on-chain data will show whether merchant adoption of PYUSD for settlement is accelerating alongside the agentic narrative or lagging it. Third, watch the Senate floor: if the GENIUS Act or CLARITY Act framework passes before Q3, expect a wave of enterprise treasury teams to begin formally scoping PYUSD and USDC for agentic spend budgets. The window to shape these standards, as the Consensus program itself noted, is narrow — and it is closing now.

Sources

Primary sources and prior BlockAI News coverage referenced in this article.

Primary sources

• Google Cloud Blog — Introducing an agentic commerce solution for merchants from PayPal and Google Cloud
• PayPal Newsroom — PayPal Brings PayPal USD to Users Across 70 Markets (March 17, 2026)
• PayPal official — Agentic Commerce with PayPal (AP2 collaboration page)
• PR Newswire — PayPal USD 70-market rollout press release (March 17, 2026)

From BlockAI News

• Stablecoin's Second Wave Has a New Engine: the AI Agent
• AWS, Coinbase & Stripe Launch AI Agent Stablecoin Payment Rails
• How AI Agents Pay With USDC: x402, EIP-3009, and Agentic Commerce
• Senate Banking Sets May 14 Vote on CLARITY Act After January Collapse
• Coinbase's Trading Model Is Dying. Its Next One Is Just Being Born.

How we report: This article cites primary sources, regulatory filings, and on-chain data where available. BlockAI News uses AI tools to assist with research and first-draft generation; every article is reviewed and edited by a human editor before publication. Read our full How We Report page, Editorial Policy, AI Use Policy, and Corrections Policy.