World says its proof of human can secure bank wires and AI agents. Five design partners include no bank, and only about 18m people have passed the check.
An employee of Arup wired about $25 million in early 2024 after a video conference in which every other participant was a deepfake. A deepfake is a fake video or voice made by AI. World now says it can prove that a real person stands behind a call, a wire request or an AI agent placing an order.
World's cofounder Alex Blania unveiled the product on stage at Money20/20, a fintech conference in Las Vegas. In a post dated 18 October, World said its suite for banking and finance was now available. The product answers one question: is a unique human behind this account or agent?
Banks lose money on two further questions. Was that person authorised to act, and who pays when an agent gets it wrong? So Proof of Human helps only a narrow set of high value steps, chiefly wire approvals and account recovery. It does not yet reach the customers of a retail bank, and the first partners sit close to the company that issues the proofs.
The suite has three parts: the World ID app, World ID for Agents and World ID for Zoom. It offers three kinds of proof. Proof of Human confirms a real and unique person after a one time visit to an Orb, World's scanning device. Proof of Presence is a quick liveness check on a phone. Proof of Attributes can show nationality or an age threshold while revealing only that fact.
For banks, the post names human binding approvals for ACH and wire transfers. ACH is the US batch payment network. It also names human binding requirements for account recovery, meant to prevent account takeover, where a criminal seizes a customer's login.
For agents, the post offers AgentKit, a set of developer tools that it says is currently in beta. With it, a person can delegate certain proofs to an autonomous agent. World's own guide describes it as a toolkit for human backed agents that works with agentic payment protocols such as x402. AgentKit therefore predates the 18 October announcement.
The partner roster has no bank. The post says Bain Capital, OpenExchange, Outtake, Pantera Capital and VanEck Funds are initial design partners utilizing the technology. It describes no deployment.
Two of the five already have a financial link to World. According to World Foundation's release, Pantera Capital led a $52.5 million raise of locked WLD tokens for World Foundation on 24 July 2026. Bain Capital Crypto, the crypto arm of design partner Bain Capital, also took part.
Token buyers hold no equity or profit rights in Tools for Humanity, the company that builds the Orbs. So they are token holders and not owners of the business.
Design partners are how every identity product starts, and the release does say "initial". At that stage the roster is reasonable. A deposit taking bank signing would change the picture. A bank carries the customer losses, the regulator and the vendor review that an asset manager does not.
Coverage is the second limit. The post says more than 18 million people hold a Proof of Human. Bitcoin.com put the network at more than 39 million members in July. So about 46 percent of people who joined have been scanned.
The count has barely moved. The same July report gave the verified figure as 18 million, so it has been described as 18 million since then. In May 2025 it was 12.5 million, so the base grew about 44 percent over 17 months.
A retail bank has millions of customers, and each would have to visit an Orb. The post says Orbs are currently available at the NYSE. That is a conversion cost the bank pays before any fraud is avoided.
The liability gap is the harder problem. Suppose a client delegates a proof to an agent and the agent buys the wrong security. The proof shows that a person delegated it. It does not show that the trade was within the mandate, under the limit or consistent with what the client intended.
A bank still needs a record of the agent's authority and caps on size and counterparties. It also needs a rule for who bears the loss when the agent errs. Until those exist, human binding for agents settles who is behind the order. It leaves who pays untouched.
Fraud is rising, and the product is designed for that problem. The post cites Deloitte as predicting that AI enabled fraud losses in the United States could reach $40 billion by 2027. Deloitte's estimate for 2023 was $12.3 billion.
Deloitte puts that at a compound annual growth rate of 32 percent, meaning losses grow by about a third each year. The direction is not in doubt.
The Arup fraud worked because a person was tricked by faces and voices that looked right. The transfer was approved by someone who believed the meeting was real.
World ID for Zoom shows a Human Badge on the participant tile. The post says hosts will soon be able to require Proof of Presence in a waiting room. That check stops impostors only when both sides join the call with the badge.
A scammer who has no badge is easy to spot. But a scammer who persuades a real employee to ignore the missing badge faces no barrier at all.
The practical use is narrow. Proof of Human fits as a step up for high value wires and account recovery among customers who already walk into a branch. There, one Orb visit is a tolerable ask. The cost is friction and the exclusion of anyone without a scan.
Leaving recovery on documents that deepfakes already defeat carries its own risk. FinCEN, the US Treasury's financial crimes unit, warned about that in 2024. Waiting for an industry standard on agent authority may take years.
A regulated deposit taking bank could turn five design partners into a market. It would need to become a live user for customer wire approvals by 31 March 2027. If the next announcements still come from asset managers, crypto firms and security vendors, the 18 million ceiling is the product.