World ID 4.0 Wants to Be the Internet's Proof You Are Human. Read the Fine Print.

The internet is getting a layer that tells humans apart from machines, run by a private company

On April 17, 2026, the Sam Altman backed project World shipped the largest version yet of its identity protocol, World ID 4.0, with a claim that should make every security leader look up. It says it can confirm that a real, unique human, and the same one each time, is on the other end of a digital interaction, without learning who that person is. The timing is not subtle. Deepfakes now pass in live video calls, and AI agents have started to click, buy, and sign on people's behalf.

World calls the result full-stack proof of human. Strip away the launch language and it is an attempt to build a missing layer of the internet: a reliable way to tell a person from a machine. The engineering is genuinely interesting. The questions it raises are bigger than the engineering.

What actually launched

Proof of human (POH) is the core idea: a credential that proves you are a unique human being, not a bot or a synthetic identity, while revealing nothing else about you. The credential is minted by an Orb, World's iris-scanning device, which photographs your eyes, turns the texture of your iris into a cryptographic code called an IrisCode on the device, and then checks that code for uniqueness against a blockchain database using a zero-knowledge proof (a method that proves a fact, here you are new and unique, without exposing the underlying data). The original images are deleted by default, and the code is split into fragments across separate servers so no single store holds a usable copy.

Version 4.0 is the re-engineering around that core. It adds key rotation (so a stolen key can be swapped without losing your identity), multi-party entropy and one-time-use nullifiers (cryptographic tricks that stop two of your interactions from being linked together), recovery, and formal session management, the unglamorous features enterprises require. The headline addition is what World calls human continuity: proof that the same real person is present across separate interactions. Today's systems verify a device or an account, not a human, so whoever holds your laptop or your hardware key can act as you. World also shipped a Selfie Check that verifies without an Orb, agent delegation tools it likens to a power of attorney for your AI assistant, and a World ID app.

• Nearly 18 million people across 160 countries have verified at an Orb, and World says credentials have been used hundreds of millions of times.
• Launch partners are not fringe: Zoom will badge verified humans on calls to fight deepfakes, while Okta, Docusign, and Tinder are integrating proof of human into login, contracts, and dating.
• World ID for Agents aims to bind each AI agent to a verified human standing behind it, the protocol's answer to an internet filling with autonomous software.

Why a security leader should care, and stay skeptical

The defensive case is real. If a deepfake can join your finance team's video call, a per-participant proof of human badge is a meaningfully harder thing to fake than a face. As agents proliferate, knowing a verified person authorized an action is a control worth having. This is why the integrations matter: proof of human is moving from a crypto curiosity into mainstream identity and access management, the layer your organisation already lives in. But the criticism is just as substantial, and it comes from serious people. Edward Snowden and security practitioners warn that concentrating biometric personhood inside one private company creates a central point of failure and an unaccountable gatekeeper. David Shipley of Beauceron Security contrasts the model with Apple's, where biometrics never leave the device, and argues proof of personhood is a public good that should not be sold by a private firm. There is a harder technical problem underneath: an iris, like a fingerprint, cannot be reset. If the link between a person and their credential is ever broken or stolen, you cannot reissue the eye. World's answer is heavy cryptography, anonymization, on-device custody, and key rotation, which addresses the credential layer but not the permanence of the biometric itself. The systemic shift to weigh is this: the AI era is creating real demand for a human layer of the internet, and whoever supplies it holds enormous leverage over who gets to participate online. Several governments have already pushed back, with World banned or suspended in Kenya, Brazil, Indonesia, Hong Kong, and Spain over data-protection concerns.

What to do about it

1. Treat proof of human as one signal, never as your whole identity stack. Wire it in beside existing authentication and authorization, because a single external personhood provider is exactly the central dependency your architecture should avoid leaning on alone.
2. Demand the data-handling evidence before you adopt. Ask where the biometric is processed, what is deleted, and whether independent audits exist, and remember a leaked biometric is permanent in a way a leaked password never is.
3. Do not confuse a verified human with an approved action. For agent delegation, keep an explicit human-in-the-loop step for payments and high-impact decisions, because proof that a person stands behind an agent is not proof that the person reviewed what the agent just did.

Bottom line

World ID 4.0 is the most serious attempt yet to solve a problem the AI era made urgent: telling humans from machines online. The cryptography is real and the deepfake-resistant use cases are compelling. The open question is not whether the technology works, but whether the internet's proof of personhood should sit inside one private company, secured by a biometric no one can ever change. Evaluate it as a useful signal, not a foundation, and read World's own announcement alongside the privacy critics before you wire it into anything that matters.