Stripe has long ceased to view itself as a payment company; it now positions as an infrastructure provider for the AI economy. A deep-dive analysis argues a counterintuitive thesis: the next-generation payment model will not be designed from the payment layer, but derived from the Agent economy, making payment only a subsystem of KYA.
The Five-Layer KYA Architecture: Identity, Intent, Credit Sit Outside the Payment Rails
KYA (Know Your Agent) is not a payment-layer product; it's the infrastructure layer for the Agent economy. The analysis defines five KYA layers — Agent identity, authorization scope, intent signature, liability chain audit, and credit rating — of which only authorization scope and liability chain audit fall on the payment rails. The other three layers (identity, intent, credit) sit entirely outside the payment domain.
- Identity layer serves all scenarios requiring Agent recognition: cross-platform calls, regulatory filing, internal audits — payment is just one use case.
- Intent layer serves the larger problem of AI alignment — payment is only one verification scenario among many.
- Credit layer serves any system that needs to assign permissions and limits to Agents — payment is again just one consumer.
One industry observer claimed that "the only change is payment and settlement efficiency," essentially framing KYA as a subsystem of payments. But the analysis flips it: payment is actually a subsystem of KYA.
Hard Data: The Real Cost of an AI Transaction Occurs Before Payment
Stripe's data head Emily Glassberg Sands shared numbers in an Every interview: a large AI client blocks 250,000 fraudulent free trials per week; one AI company burns $25 in computing power per free trial with only a 4% conversion rate, meaning each abusive user costs the company 625x the revenue of a legitimate user.
These figures prove that in the AI economy, the decision of whether a transaction should happen is no longer made at checkout. It's made upstream, in questions like "who is this, what do they want, and does it deserve resources?" That's why Stripe moved its Radar risk engine from "transaction moment" to "full user lifecycle": not to speed up old risk control, but to shift the question from "is this payment suspicious?" to "is this user/Agent's entire behavior suspicious?" The former is a payment-layer question; the latter belongs to KYA.
Stripe's Bet: Acquisitions of Bridge, Privy and Building Tempo Blockchain
At Sessions 2026, Patrick Collison didn't say "AI payments." He said "economic infrastructure for AI." Not a marketing phrase — a strategic claim. Stripe co-built the Agentic Commerce Protocol (ACP) with OpenAI, now adopted by Microsoft Copilot, Meta, and Google Gemini. It's essentially an identity and session protocol, not a payment protocol. Shared Payment Token isolates Agents from real card numbers — an authorization-layer move, not a settlement-layer one.
Stripe acquired Bridge for stablecoin infrastructure, Privy for embedded wallet capability, and self-built Tempo blockchain for settlement rails. None of these moves fit in the "payment efficiency optimization" frame. This portfolio only makes sense under the premise that KYA is the foundational layer. If the Agent economy were merely a payment efficiency problem, Stripe wouldn't need stablecoins, embedded wallets, or its own L1. What it's doing is occupying each of those five KYA layers.
Liability Chain Goes Networked: KYA Turns 'Who to Blame' from Unverifiable to Verifiable
The ultimate legal subject remains a human — AB 316 has codified that. But the real problem KYA solves is this: when the liability chain becomes distributed, figuring out exactly which step on which human is responsible becomes a task that KYC didn't need but KYA must handle.
In the KYC era, the liability chain was linear (user → payment/bank → merchant). If a transaction goes bad, you instinctively know whom to call. In the KYA era, the chain is distributed (user → Agent platform → model provider → payment protocol → bank → merchant, with possible inter-Agent calls). Even if the law says "sue the human, not the Agent," you still don't know which human — because responsibility is scattered across 5 to 7 parties.
KYA cannot change the ultimate legal liability. But it can cryptographically record each party's role and action in the distributed chain — who authorized what, who executed what, who settled what, who performed what. It turns "no evidence" into "evidence available"; it turns "which step went wrong is unverifiable" into "verifiable." This isn't payment efficiency. It's the first time accountability becomes traceable in an Agent network.
What exactly is the next-generation payment model? Still unclear. But one thing is certain: it won't be designed within the payment layer. It will emerge from the scenarios after KYA infrastructure is laid.