World Pushes Proof-of-Human IDs into Banks, Allows Delegation to AI Agents

Sam Altman-backed World (formerly Worldcoin) is pushing its proof-of-human identity credentials into banking channels and allowing users to delegate their identity proofs to AI agents. According to Decrypt, World cites a $40 billion AI fraud risk as the rationale for this move.

World's core product is an iris-scan-based human identity verification system: users complete a scan via an Orb device and receive a credential proving they are "a real and unique person." Previously, this credential was mainly used in World's own wallet and token distribution scenarios. What's different this time is the simultaneous extension in two directions—toward regulated financial institutions such as banks, and toward AI agents.

Moving toward banks means World is attempting to bring proof-of-human from crypto-native scenarios into traditional finance's KYC processes. Banks have always had strict identity verification requirements, but existing KYC solves "who is behind this account," not "is the person initiating this operation actually this person." As AI agents begin to initiate payments, transfers, and contract signings on behalf of humans, banks face a new question: how do you confirm that an instruction initiated by an agent has a real, authorized human behind it? If World's credential is accepted by banks, it effectively adds a layer of "human-ness" proof on top of traditional KYC.

Moving toward AI agents means allowing users to "delegate" their identity proof to agents. This design directly touches the most thorny aspect of agent payments: authorization. For an agent to move money, it must prove two things—whom it represents, and what it is permitted to do. World's approach is to have a real person first prove their existence, then grant the agent the right to use that proof. Formally, this resembles an authorization structure of "human endorsement + machine execution."

The $40 billion AI fraud risk figure is the story World is telling banks and regulators: AI-generated content, deepfakes, and automated accounts are making it increasingly difficult to judge "is the other side human," and fraud losses are accumulating. According to Decrypt, World uses this as the reason to push proof-of-human into mainstream finance. The figure itself comes from World's citation; the specific methodology and calculation were not detailed in the report, but the problem it points to is real in the industry—once agent payment volume picks up, the identity verification gap will be rapidly amplified.

From the perspective of agent payment compliance and security, the significance of this development lies not in whether World's iris-based solution itself will be widely adopted by banks, but in the fact that it formally places the "identity layer" as a prerequisite for agent settlement. Currently, industry discussions about agent payments focus mostly on the protocol layer (such as x402, MCP) and the settlement layer (stablecoins, on-chain clearing and settlement). But when an agent initiates a payment, what the payee and payment service provider really need to answer are three questions: who is behind this agent (KYA, Know Your Agent), what is it authorized to do (authorization intent, per-transaction limits, payee whitelist), and should this transaction be allowed (pre-settlement interception). World's proof-of-human attempts to answer half of the first question—is there a real person behind it; the delegation mechanism attempts to answer part of the second question—has the real person delegated authority to this agent.

But there is a structural gap here. Proving "there is a real person behind this" and proving "this agent's current operation is within the authorized scope" are two different things. The former is identity; the latter is continuous verification of authorization intent. A real person can prove their existence and then delegate the credential to an agent, but the boundaries of that delegation—how much can be paid, to whom, how many times per day—are not automatically constrained by proof-of-human. In other words, what World provides is an identity-layer primitive, not a pre-settlement risk control layer. If banks only integrate the former without the latter, they still cannot intercept an unauthorized agent payment before settlement.

For banks, World's proposal is both attractive and costly. The attraction is that the pressure of AI fraud is real-time, and banks need a means to quickly distinguish humans from machines; the cost is that incorporating iris credentials into the KYC system involves compliance of biometric data, cross-border data flows, and how responsibility is attributed after credentials are delegated. According to Decrypt, World is pushing this solution to banks, but the report did not disclose the list of banks that have signed on or specific cooperation terms.

For other participants in the agent payment ecosystem—payment service providers, stablecoin issuers, wallets, and custodians—the signal is: identity verification is shifting from an "optional risk control plugin" to an "access condition for settlement channels." When banks begin to require agent payments to carry proof of human identity, agent wallets and PSPs without an identity layer will encounter friction in deposit and withdrawal processes. Conversely, whoever can link identity proof, authorization intent, and pre-settlement interception into a single chain will be closer to agent payment infrastructure acceptable to financial institutions.

Whether World's solution can become a standard depends on whether banks are willing to accept a non-traditional biometric credential, and whether regulators recognize the legal validity of the "delegation to agent" authorization form. Neither question currently has an answer. But pushing proof-of-human from token distribution scenarios to banks and agents at least demonstrates one thing: for agent payments to move toward real money, the identity layer is an unavoidable prerequisite.

---

Source: Decrypt · https://decrypt.co/381288/sam-altmans-world-takes-proof-of-human-ids-to-banks-citing-40b-ai-fraud-risk