AI Agents Can't Open Bank Accounts. Google and PayPal Say Crypto Is the Clear Answer.

13 May 2026 5 min read

Google Cloud's global head of Web3 strategy stood on stage at Consensus Miami last week and said something that should change how you think about cryptocurrency. "An agent cannot get a bank account," Richard Widmann told the audience. "It's not hard. It just is impossible." PayPal's crypto chief, May Zabaneh, agreed — telling the same conference that the company's stablecoin PYUSD is "a very natural programmable layer" for the AI agent economy that's already arriving. Google has launched a new protocol called AP2 (Agent Payments Protocol), donated to the FIDO Foundation, with more than 120 partners including PayPal. The protocol is designed to let AI agents make and receive payments autonomously — using stablecoins and crypto rails instead of bank accounts they'll never be able to open.

Three things to know:

AI agents are already showing up as customers. A PayPal survey found that 95% of merchants are already seeing AI agent traffic on their platforms — bots browsing, comparing, and in some cases purchasing on behalf of human users. But only 20% of those merchants have machine-readable product catalogues, meaning the infrastructure for agent-driven commerce barely exists yet. Qualcomm's CEO has called 2026 "the year AI agents go mainstream." The agents are arriving. The payment rails they need are not ready.
Traditional banking is structurally incompatible with AI agents. A bank account requires a human identity. It needs KYC verification, a physical address, a signature, and in many countries an in-person visit. An AI agent has none of these things. It has no passport, no utility bill, no face to present at a branch. It operates at machine speed, executing hundreds of transactions per second across multiple jurisdictions simultaneously. Traditional banking was built for humans doing business during office hours. AI agents operate 24/7, globally, and programmatically. The mismatch is fundamental.
Crypto wallets solve the problem that banks can't. A crypto wallet can be created in milliseconds by a line of code. It doesn't require identity verification to hold or transfer value. It settles instantly, works across borders, and can be programmed to execute complex payment logic — escrow, conditional release, multi-party splits — without human intervention. This is why Google and PayPal are building AI agent payment infrastructure on crypto rails rather than trying to force agents into traditional banking. The technology was designed for exactly this kind of use case, even if nobody predicted it ten years ago.
Compliance is the real bottleneck. Crypto wallets solve the payment problem, but the business on the receiving end still needs to know who it's dealing with. Anti-money laundering rules, consumer protection laws, and tax reporting all assume there's a human behind every transaction. An AI agent can't hand over a passport. Google's AP2 protocol is working on this — letting agents present verifiable credentials on behalf of their human owners, like a digital power of attorney. But most jurisdictions haven't defined how agent-initiated transactions satisfy compliance requirements. The technology for AI payments is ready. The compliance infrastructure is not.

The Economy That's Coming

Think about what AI agents already do. They book flights, write code, manage calendars, answer customer service queries, analyse documents, and generate reports. The next step — already happening in 2026 — is agents that can spend money on your behalf. An AI agent that monitors your electricity usage and automatically switches to the cheapest provider. One that negotiates your insurance renewal by comparing quotes across twenty providers in seconds. One that orders office supplies when inventory drops below a threshold.

Each of these actions involves a payment. And each payment currently requires a human to authorise it through a banking system designed for human interaction. As agents become more autonomous, that bottleneck becomes untenable. You can't have an AI agent that negotiates a deal in four seconds and then waits three business days for a bank transfer to clear.

Stablecoins fix this. A USDC or PYUSD transaction settles in seconds, costs fractions of a cent, and can be triggered programmatically by an agent without any human touching a banking app. The agent earns crypto, spends crypto, and settles crypto — all within the same programmable infrastructure.

Google's AP2 protocol standardises how this works. It defines how agents identify themselves to merchants, how payments are authorised and settled, and how disputes are handled — all without assuming the payer has a bank account. The 120+ partners building on AP2 suggest this is going to be an industry standard, not a niche experiment.

The Compliance Bottleneck

Crypto wallets solve the payment problem, but they don't solve the compliance problem. An AI agent can hold a wallet and send a transaction in milliseconds. But the business on the receiving end still needs to know who it's dealing with. Anti-money laundering rules, consumer protection laws, and tax reporting requirements all assume there's a human behind every transaction. Most businesses are legally required to perform some form of KYC on their customers — and an AI agent can't hand over a passport.

This is the gap that Google's AP2 protocol is trying to bridge. The protocol includes mechanisms for agents to present verifiable credentials on behalf of their human principals — linking an agent's wallet to a verified identity without requiring the agent itself to be a person. Think of it as a digital power of attorney: the human passes KYC once, and the agent inherits the authority to transact within defined limits.

But the frameworks for this don't exist yet in most jurisdictions. Until regulators define how agent-initiated transactions satisfy compliance requirements, businesses face a choice: accept agent payments and risk regulatory exposure, or reject them and lose customers to competitors who don't. The technology for AI payments is ready. The compliance infrastructure is not. That gap is the real bottleneck holding back full adoption of automated commerce.

What This Means for South Africa

South Africa is further into AI adoption than most people realise. Over 60% of South African workers regularly use generative AI — a higher rate than the UK, France, or the United States. Microsoft has trained 1.4 million South Africans in AI skills through its partnership with SABC Plus. Ninety-three percent of SA businesses have started their AI journey in some form.

But South Africa's banking system is particularly hostile to autonomous agents. FICA verification requires proof of identity and address. Many banking services still require in-branch visits. Exchange controls restrict cross-border payments. Even human South Africans find the banking system bureaucratic. For an AI agent, it's a wall.

This is where crypto infrastructure becomes strategically important for South Africa — not as a speculative asset, but as the payment layer that lets South African businesses participate in the agent economy. A South African freelancer whose AI agent handles client invoicing, payment collection, and expense management needs that agent to transact. If the only payment rails available to agents are crypto-based, then South African businesses that can't access crypto infrastructure are locked out of the next wave of commerce.

The Draft Capital Flow Management Regulations closing for comment on 18 May propose bringing crypto under exchange controls. Those controls were designed to manage human capital flows in a pre-digital era. Applying them to the payment rails that AI agents need is like applying horse-and-cart speed limits to the internet. The technology has moved. The question is whether the regulations will move with it.

Google and PayPal aren't telling you to buy Bitcoin. They're telling you something more important: the next generation of commerce will run on programmable money, because the customers doing the buying won't be human.

Sources:

Leon Kowalski