In São Paulo, on a stage built for optimism, three institutions that spend most of their energy fighting each other agreed to pretend they don't. Visa. Mastercard. Ant International. On the same day, in the same room, they launched something called Know Your Agent — KYA — a shared identity framework for the AI agents that will soon shop, book, and pay on our behalf.
The data they put on the screen was the interesting part. Not the 4,700% growth in AI-driven retail traffic — growth rates are cheap in a bull narrative. The interesting number was 14. That's the share of consumers who say they trust an AI agent to make a purchase without human verification. And of those skeptics, 42% won't let an agent near a transaction worth more than $25.
Four thousand seven hundred percent. Fourteen percent. Hold those two numbers side by side and you have the whole story.
Context
Here's what the framework actually is. KYA is not a product. It's an interoperability layer — a translation protocol that sits above three proprietary identity stacks: Visa's TAP, Mastercard's Verifiable Intent, and Ant's AMP. Register an agent once and, in theory, every network recognizes it. One identity, many rails.
The architecture matters less than the politics. Visa brought twelve launch partners — Adyen, Shopify, Stripe among them. Mastercard brought Google, co-building the Verifiable Intent spec as open source. Ant brought the thing neither card network can manufacture: 1.3 trillion dollars of wallet-native spend across emerging markets where traditional credit penetration is thin and digital wallets are the default.
Three protocols. Three fiefdoms. One claimed passport.
The bet underneath it isn't about identity at all. It's about the pie that arrives when autonomous agents start transacting at machine speed. McKinsey puts agentic commerce at three to five trillion dollars by 2030. The framework is a way to pre-build the turnstiles before anyone knows where the stadium is. The efficiency argument — higher operating efficiency, lower barriers for agents to enter — is real. It's also a polite way of saying: we intend to be the toll booth.
And note what's absent. No regulator spoke at the launch. No sandbox, no license consortium, no supervisory approval. Just three firms deciding the rules first and asking permission later. When private infrastructure sets identity standards ahead of public law, it is quietly writing regulation — and calling it a specification.
The choice of São Paulo as launchpad is itself a signal. Emerging markets, where wallets skip credit cards entirely, are the lowest-friction test bed for agent payments — and the place where a wallet challenger carries native leverage.
Core
Begin with the hard part, because the press release won't. The framework's central unresolved question is not protocol translation. That's solvable engineering. The hard part is the trust root.
Who maintains the unified registry of agent identities? Who issues the certificate that says "this is a legitimate agent acting for this human"? The answer is buried under the word "interoperability," and it's buried on purpose. Because whoever runs that registry holds something structurally similar to a root certificate authority — the power to issue, revoke, and set pricing for the fundamental unit of trust in machine commerce.
I've spent time inside identity and authentication systems, tracing how a single registry becomes a systemic chokepoint. Based on that work, I can tell you the failure that nobody is modeling: if a cross-network agent registry gets attacked or goes dark, agent transactions across all three rails stall simultaneously. KYA's genius is its efficiency. Its blind spot is that it concentrates trust risk — previously spread across separate networks — into one new, undefined component. That's efficiency traded for fragility, and the trade is being made before the governance exists to manage it.
Then there's the verification problem, which is deeper and more interesting than identity. Cryptography can prove which agent is transacting. It cannot prove the transaction reflects a human's actual intent. Mastercard's partnership with Google on Verifiable Intent is aimed precisely here — at the semantic layer, at proving not just identity but purpose. That's the sharper moat, and everyone in the room knows it. Identity is the gate. Intent is the brain. Whoever defines intent verification defines the decision layer of agentic commerce.
Now watch the money move. In 2025, Visa acquired BioCatch — behavioral biometrics — for 2.4 billion dollars. Read that as evidence of where the risk actually lives. Behavioral biometrics are sensitive personal data under GDPR and comparable frameworks. Embedding them into an agent identity chain means using a fragment of your body as the trust anchor for a machine's authority. The compliance cost just went up. The breach consequence went up more.
So the three players are building layer by layer: Visa buys the human signature, Mastercard builds the intent protocol, Ant supplies the distribution. Look at what each really controls.
Visa and Mastercard are defending a position nobody has attacked yet. Their presence isn't expansion — it's insurance. If agent commerce arrives, they want the agent to arrive through a Visa or Mastercard turnstile. And here is the quiet asymmetry: when an agent transaction is recognized across three networks, it can be routed competitively. Cross-network identity recognition ultimately erodes the card networks' historical lock on routing. Which is why their apparent openness at the identity layer masks a defensive move at the clearing layer. They are trading a controlled open identity standard to keep the clearing default intact.
Ant's position is the most complicated and, to me, the most instructive. Ant holds the strongest distribution edge here — a wallet ecosystem the card networks cannot replicate, especially across Southeast Asia and Latin America. But distribution without standards authorship is a fragile kind of power. In a framework where identity and intent are defined by others, the biggest network can still become a subordinate endpoint. Ant joins KYA to gain a compliance passport for cross-border expansion — and risks becoming a translator for someone else's protocol, valuable but not sovereign.
There's another structural gap the analysis keeps circling: the banking layer. KYA lives at the agent-network-merchant edge, but money still has to settle into real accounts. Without issuer and account-authorization integration, "interoperability" stops at front-end authentication and never becomes an end-to-end settleable loop. The moment one of these executives says "issuers," as one did in São Paulo, they're signaling that the next phase must drag banks inside.
The unit economics tell their own story. Registration-once-use-everywhere cuts marginal onboarding cost for developers and merchants. Good. It does almost nothing for the cost of consumer trust. And trust is where the revenue bottleneck actually sits. The $25 threshold isn't arbitrary. It's the price at which chargeback economics flip — below it, disputes cost more than they're worth, so the system absorbs fraud and pretends it isn't happening. Above it, liability is unassigned and the machine stalls. The $25 line isn't a behavioral quirk. It's the exact point where the framework's missing liability rules become visible.
Which is why the model underneath KYA is best described as rent deferred. Nobody charges for identity today. The plan is to grow the total volume of agent commerce and tax the clearing layer later. Identity is the free water. The transaction is the meter. But the profit split between the identity layer, the clearing layer, and the distribution layer isn't defined — and in every undefined standard, the party who contributes the most traffic often discovers it has the least pricing power.
Contrarian
Here's the angle the launch wants you to miss. The real competition in agentic commerce is not Visa versus Mastercard versus Ant. Those three already decided to stop fighting long enough to build a fence. The threat that pushed them together is upstream: the platforms that own the agent itself.
If OpenAI, Google, or Apple build identity and payment directly into the agent runtime, the agent may never "touch" the card network's identity layer at all. KYA becomes a backend query the platform makes or skips. That's the genuine battle — card consortium versus agent platform, fought over who sits closest to the user's intent. Mastercard inviting Google to co-build Verifiable Intent is the defense of a vulnerable incumbent turning a potential disruptor into a partner. It's clever. It's also how you let a wolf design the stable door.
And there's a second blind spot nobody at the launch wanted to name. By extending KYC and AML from people to agents, KYA creates a new kind of financial subject — an entity with a verified identity but a genuinely ambiguous soul. When a hijacked agent executes a transaction across three networks, who files the suspicious activity report? The initiator, the network, or the registry? The interoperability that makes the system elegant is the same interoperability that lets responsibility evaporate between three jurisdictions. The framework's greatest strength — that it spreads one identity across many rails — is also its greatest liability: it spreads one failure across many parties until no party owns it.
Takeaway
So watch two things. First, who is awarded the registry — because whoever runs it will, over time, become the closest thing to a central bank of machine identity, and it won't be a central bank. Second, whether the trust scissors-gap — 4,700% growth against 14% trust — actually converges, or just produces a spectacular volume of experiments that never become revenue.
The bear market didn't kill speculative infrastructure. It exposed which speculative infrastructure had a real load-bearing wall inside it. KYA has one — the entire economy is going to need machine identity — but it's holding that wall up with the wrong hands, in the wrong order, before the regulation that will eventually rewrite it has even shown up.
We don't get to decide whether machine identity becomes centralized or open by publishing a spec — we decide it by who we let hold the keys.
About Me: I write about why protocols fail and why the failures matter. Not because I enjoy the wreckage, but because the code that survives a bear market is written by people who studied the ones that didn't.
