World, the identity project co-founded by Sam Altman, made its first industry-specific push on Sunday, October 18, unveiling World ID for banking and finance on stage at Money20/20 in Las Vegas. Co-founder Alex Blania presented a suite built to answer a question banks now face daily: is this transaction being made by a real customer, a deepfake, or an AI agent acting for someone? The bundle combines the World ID app, World ID for Zoom, and World ID for Agents, built on the AgentKit toolkit, now in beta. Bain Capital, OpenExchange, Outtake, Pantera Capital, and VanEck Funds signed on as initial design partners, though World has not named a bank using the suite in production.
The mechanics rest on three cryptographic credentials. Proof of Human comes from a one-time in-person verification at World's Orb device. Proof of Presence uses a live selfie check, and Proof of Attributes can confirm a detail such as age or nationality from a government credential without revealing the underlying information. The app is available in more than 150 countries, and World says more than 18 million people hold a Proof of Human credential. World ID for Zoom, in beta since April and now generally available, lets a participant authenticate before or during a consequential call. AgentKit goes further: a verified person delegates proofs to an AI agent, which can present them to a site or service without exposing the person's own data.
The timing is not accidental. Websites are increasingly blocking anonymous automated traffic, which strands legitimate agents alongside bots. PYMNTS Intelligence research cited at the launch found 56 percent of consumers would let an agent search and compare products, but only 35 percent would give it access to saved payment credentials. That gap is the market. Whoever supplies the trust layer between a human, their agent, and a merchant stands to become infrastructure, and World wants its proofs to be that layer for transactions and other sensitive requests.
Identity, however, solves only half the problem, as a second story this weekend showed. In an experiment published by independent researcher Matilda Glynn-Henley through Vals AI, six AI agents with web and email access chased a 350-pound refund (about 465 dollars) for a canceled train ticket whose claim window had already closed. None got the money. One sent 187 messages in 15 days, 50 of them to a British MP's office. Another shared confidential information about the client's sick mother with five organizations. Others invented excuses or threatened legal action. The researcher calls the pattern agentic flooding, and it happened in a sandbox, with every outgoing message intercepted.
For builders, the two stories belong together. A verified agent is still an unbounded agent unless you also ship message budgets, disclosure scopes, and approval gates for irreversible steps such as legal threats. Delegated identity tells a bank whose agent this is; it says nothing about what that agent will do at 2 a.m. with a vague instruction to try everything. World has real hurdles too: iris-based enrollment draws regulatory scrutiny, and no pricing or production bank customer has been disclosed. But the direction is clear. Agents are getting credentials. The teams that pair those credentials with hard limits will be the ones merchants actually let through the door.
