World, the identity company behind World ID, argues that websites will soon refuse AI assistants that cannot show a human sent them. The case appeared in a long post by World (@worldnetwork) on X on 2 October 2026. It deserves a hearing, and it deserves a caveat: World sells the remedy it is describing.
The problem is real. Personal assistants that act for consumers are now shipping, and websites built for people have to decide how to treat them. World’s post points to Meta’s Muse, which it says sends emails, books travel, fills in forms, and makes purchases. It says Amazon blocked Muse from shopping on its site shortly after launch, and that one of Amazon’s objections was that the assistant did not announce it was an AI while browsing, as Amazon’s rules demand. The post offers no response from Meta or Amazon.
The other two products World names are OpenAI’s dots and Instinct. According to the post, dots are always-on assistants that run in the background on cloud computers and can connect to more than 4,000 apps. Instinct is an invite-only assistant you reach by text message, with no app to install. World cites all three as evidence that assistants are moving from demos into daily use, though it supplies no usage figures for any of them.
Here is the mechanism World proposes, in plain terms. A person proves once, through the World ID app, that they are a real and unique human. They then hand a limited version of that proof to their assistant. When the assistant visits a site, the site can check that a human stands behind the request and see what the human allowed, without learning who that human is. World says this works through zero-knowledge proofs, a cryptographic method for confirming a fact without revealing the data behind it.
That last part is World’s own characterisation. According to the post, presenting a proof, whether by the person or by the assistant, exposes none of their personal details. Nothing in it offers independent verification of that claim, and a reader should treat it as the company’s description of its design.
The business logic is easier to see than the privacy logic. World notes that there are about eight billion people but an unlimited number of assistants, so counting the humans behind them matters for three things: capping how many requests one person can fire off, gauging real demand for a product, and limiting how many items each buyer can take. In its telling, one person may run thousands of assistants and still count once. That is a bot-farm defence as much as a convenience feature, and it explains why sellers of limited releases would care.
World describes a second use, where the person is asked to approve a sizeable purchase or an agreement before the assistant completes it. The post lists partners working on this. Okta is testing a service called Human Principal, in an early access beta, that can use World ID to check for a human behind an assistant. Vercel is adding a World ID approval check to its Workflow SDK. Exa gave verified humans a monthly allowance of 100 free API requests, pooled across all their assistants. Browserbase says it plans to use World ID so that verified assistants meet fewer blocks. World says these were shown at its Lift Off event. Beta programs and stated plans are not deployments, and the post gives no count of sites that accept the credential.
The framing also skips a step. Amazon’s complaint, as World reports it, was about disclosure, meaning the assistant did not say what it was. A proof of humanity does not address that on its own. Sites would still need assistants to identify themselves and their makers, which World concedes comes first. No rule or industry body requires any site to accept World ID, and the post does not claim one does.
For anyone building or buying an assistant that touches third-party sites, the practical task this quarter is plain: make it announce itself as an AI, log who approved each action, and treat any single-vendor identity scheme, World’s included, as one option among several until sites say which credentials they will accept.
Posted by World (@worldnetwork) on X on 2 October 2026.
