Sam Altman’s World Taps Coinbase’s Open Protocol to Verify Humans Behind AI Agents

• World has introduced AgentKit, a developer toolkit that lets AI agents prove they represent a unique human.
• The system integrates World ID with the x402 protocol developed by Coinbase and Cloudflare.
• Platforms could use the technology to limit bots, scalpers, and automated abuse online.

World is launching a developer toolkit designed to help websites verify that AI agents are acting on behalf of real people.

Launching today in beta, AgentKit integrates Worlds World ID identity system with x402, an open protocol started by Coinbase and Cloudflare.

The system allows AI agents to carry cryptographic proof that they represent a unique human while interacting with websites, APIs, and online services.

The release aims to address an ongoing problem: AI agents taking on tasks once handled directly by users, including booking reservations, buying concert tickets, accessing APIs, and comparing prices.

“Right now, there are a lot of services where agents can spam them—social platforms, or things like ticket sales,” DC Builder, a research engineer at the World Foundation, told Decrypt. “Think of Ticketmaster: if you delegate an agent the ability to book tickets, you can spawn like 100,000 tickets. ”Even though they have the money to pay, its not a great user experience for people competing with bots.

Platforms increasingly struggle to distinguish legitimate automated activity from large networks of bots.

Earlier this month, a federal judge issued a court order against AI developer Perplexity, blocking the companys Comet browser from making purchases on Amazon on behalf of users.

AgentKit allows people who have verified their identity using Worlds orb device to delegate their World ID to an AI agent.

According to World, its network includes nearly 18 million verified individuals across more than 160 countries. While the orb rewards users with crypto in the form of Worldcoin (WLD) tokens, according to Builder, tokens arent necessary to use the AgentKit.

Once the AI agent is linked to a user‘s World ID, it can cryptographically prove that it represents a human without revealing the person’s identity.

By extending to the x402 protocol, sites can request proof alongside or instead of micropayments before granting access to services or APIs.

“What this lets you do is program against the knowledge of whether the request is coming from a human or an agent—or an agent tied to a human,” Erik Reppel, head of engineering for Coinbase‘s developer platform, told Decrypt. “As the seller, you can just say, ’This doesn‘t have proof of human attached to it, so I’m going to reject the payment.”

Platforms can then enforce limits or allocate resources based on the number of unique people interacting with a service rather than the number of automated agents involved.

“With proof of human, you at least know that the account is controlled by one person, and that there aren‘t thousands of accounts all trying to purchase something,” Builder said. “But it’s not necessary for the identity part to contain information about the individual themselves—were purely anonymous in the proof‑of‑human protocol.”

“I think there‘s a lot of the internet where it doesn’t matter that much if it‘s human or agent, and a lot of the internet where it really, really matters if it’s a human or an agent,” Reppel added. “What we need are robust, open ways of understanding which is which—being able to tell when you‘re talking to an AI, a human, or a specific human’s AI.”

In February, Coinbase introduced a wallet for AI agents on its Base network, designed to allow automated software to handle payments while keeping private keys isolated in trusted execution environments.

Beyond transactions, Builder said the technology could help maintain human interaction online as automated agents spread across the internet.

“People go on social networks to have that human connection,” Builder said. “If they want to interact with an agent, they go to ChatGPT, Claude, or another service.”