Sam Altman just announced AgentKit, a toolkit that lets verified humans delegate their World ID to AI agents. Mastercard open-sourced Verifiable Intent. Anthropic embedded Figma, Canva, and Amplitude directly into Claude mobile.
Three announcements. Three companies. One invisible infrastructure gap that's about to become visible.
The Payment Infrastructure Already Exists
Agent payments are solved. Stripe's Machine Payments Protocol repurposed HTTP 402 into a full authentication scheme for autonomous machine-to-machine payments. MoonPay's Open Wallet Standard gives agents a single encrypted vault. Ramp and Visa are issuing agent-specific credit cards.
An agent can now:
- Hit a paid API endpoint
- Receive a 402 challenge
- Fulfill it with a crypto deposit or fiat token
- Receive a cryptographic receipt
No accounts. No API keys. No human in the loop. The toll booth is built.
The Identity Infrastructure Doesn't
But here's what nobody mentions: every one of these payment systems assumes you already know which agent is authorized to spend.
You don't.
Traditional fraud signals - device fingerprints, login patterns, typing cadence - tell you nothing about machine actors. A prompt injection attack can trick an agent into buying $500 in gift cards. A hallucinated intent can trigger purchases nobody approved. From the issuer's perspective, every transaction looks identical: API call, virtual card, instant checkout.
Accenture found that 85% of financial institutions say their current systems can't handle high-volume agent-driven payments. Not because they lack payment rails. Because they lack the identity layer.
What AgentKit Actually Solves
World's AgentKit tackles this directly. One human, one cryptographic credential, privacy preserved through zero-knowledge proofs, 18 million verified users already in the network.
Integrated with Coinbase and Cloudflare's x402 micropayment protocol, it gives platforms a way to distinguish your personal shopping agent from a coordinated swarm of a thousand bots pretending to be a thousand people.
The use cases are immediate:
- Resy blocks reservation scalpers by enforcing one reservation per verified person
- Ticketing platforms stop bot armies
- Free-trial abuse disappears
The Trust Stack Is Coming Together
Layer by layer, the agentic economy's infrastructure is emerging:
- Payment rails - Stripe MPP, Visa Agent Cards, Ramp AgentCard
- Identity verification - World AgentKit, Verifiable Intent proofs
- Fraud detection - Oscilar's autonomous risk agents for non-human patterns
But there's a structural tension. The same company building the agents (OpenAI/Anthropic) shouldn't be the same company issuing the identity passports. Kenya already ruled World's biometric collection illegal. Spain ordered data deleted. If AgentKit scales to billions of agent interactions, the question isn't whether proof-of-personhood becomes essential infrastructure. It's whether a single company should be the one issuing the passports.
What to Watch
Two signals matter over the next two quarters:
First, whether identity verification becomes a requirement baked into agent payment protocols rather than an optional add-on. The first protocol that mandates "prove you're authorized by a verified human" wins the trust game.
Second, whether competing identity standards emerge. WeChat-style lock-in is valuable. But so is an open identity layer that no single company controls.
The companies building fraud infrastructure for machines, not against them, own the rails of whatever comes next. The payment layer was the easy part. The identity layer is where the real architecture challenge lives.
The difference between "agents can spend" and "we know which agents are allowed to spend" is the difference between a payment system and an economy.
Top comments (0)