The most quoted sentence at Money20/20 this year was not about payments. It was about personhood. On stage in Las Vegas, World co-founder Alex Blania unveiled World ID for banking and finance and made a claim that would have sounded absurd two years ago: “We would need a proof-of-human, and such a technology would turn out to be kind of the gate for almost any transaction on the internet” (Biometric Update).

The suite World announced on October 18 is its first industry-specific push, and it is aimed directly at agentic commerce. Three cryptographic proofs: Proof of Human, which establishes that someone is a real and unique human via a one-time iris scan at an Orb; Proof of Presence, a phone-based liveness check; and Proof of Attributes, which confirms a fact from a government credential, such as age or nationality, without revealing anything else. Around them sit the World ID app, now downloadable in more than 150 countries, a Zoom integration that puts a Human Badge on verified participants in high-stakes meetings, and most consequentially, AgentKit.

AgentKit is the piece that matters for shopping. Now in beta, it lets a person delegate World ID proofs to an AI agent. The agent can then present those proofs to any participating website or service, proving it acts at the behest of a real human without exposing that human’s personal information. As PYMNTS summarized, “that offers a way to distinguish an AI agent connected to a verified human from anonymous automated traffic that some sites are now blocking.”

That sentence describes the exact fault line running through e-commerce this month. One side is blocking: Amazon is locking AI agents out of its funnel, and several large US banks are restricting third-party agent access because they cannot tell which agents a customer actually authorized, Axios reported. The other side is opening, and World is selling the key. Both sides agree on the underlying premise: anonymous automated traffic is no longer distinguishable from hostile traffic, and identity is the filter.

Here is the problem. The identity layer now under construction authenticates the buyer. It does nothing to authenticate the buy.

The numbers that made proof-of-personhood urgent

World’s launch is well timed because the threat model it addresses stopped being theoretical this year.

  • Bots passed humans. Automated traffic surpassed human traffic on the web for the first time this year, Biometric Update notes, citing data reported by NBC News. When the majority of requests are machine-generated, “treat everyone as human until proven otherwise” stops being a workable default.
  • The fraud curve is steep. World’s announcement cites Deloitte’s estimate that generative AI-enabled fraud losses in the United States could reach $40 billion by 2027, up from an estimated $12.3 billion in 2023, a more than threefold increase in four years.
  • The Arup case is the reference point. In early 2024, an employee at engineering firm Arup in Hong Kong wired approximately $25.6 million after a video conference in which every other participant was a deepfake. The same month, FinCEN warned financial institutions that criminals were using deepfake photos and videos to defeat identity checks at account opening.
  • Scale exists. More than 18 million people already hold a Proof of Human, and World says proofs have been presented more than 475 million times. World ID 4.0, the protocol underneath, is built on zero-knowledge cryptography with post-quantum security, and the World Foundation raised $52.5 million in July in a token sale led by Pantera Capital to push deployment across consumers, enterprises, and agents.

One detail in the launch deserves more attention than it got: the named design partners are Bain Capital, OpenExchange, Outtake, Pantera Capital, and VanEck Funds. Not one bank. The institutions most exposed to agent-authentication confusion, the ones Axios reports are actively restricting agents, are watching rather than building. That gap between the fraud math and the partner list is the distance between a proof-of-human thesis and proof-of-human adoption.

The verified agent economy, and why merchants will pay for it

The most interesting part of the launch is not the identity plumbing. It is the commercial behavior World expects to unlock. Tiago Sada, World’s product chief, told Semafor that once merchants can see an agent is backed by a real human, “you can actually be much more generous.” His examples: discounts and free trials offered specifically to verified agents.

Follow that logic to its destination. A verified agent is a qualified lead with a cryptographic guarantee: a human exists, authorized this action, and can be held accountable. Compared with the anonymous bot swarm that platforms are currently blocking wholesale, a verified agent is the safest traffic a merchant can receive. Merchants will not merely admit it. They will court it, with the same perks they currently reserve for email subscribers and loyalty members.

Consumer permission data explains where the courting will concentrate. The Global Digital Shopping Index from PYMNTS Intelligence and Visa Acceptance Solutions found 56 percent of consumers would permit an agent to search and compare products, but just 35 percent would give it access to saved payment credentials. The permission boundary sits precisely where World’s delegated proofs operate: prove the human is behind the agent, and the payment gate opens. PYMNTS framed the shift precisely when it wrote that agentic commerce inserts a new question into the payments stack: not “should this transaction be authorized” but “was this machine authorized to make this decision.”

But there is a second question that nobody in this stack is asking, and it is the one that decides whether the purchase is any good: was this machine’s recommendation based on evidence that was itself trustworthy?

Proof of Human authenticates the buyer, not the buy

Walk through a transaction end to end with every new control in place.

A consumer asks their agent for the best wireless headphones under $100. The agent reads product listings, star ratings, review text, and comparison content across the open web. It shortlists, the consumer approves, the agent checks out. At checkout, World’s delegated Proof of Human confirms a human authorized this purchase. Mastercard’s probability score confirms the transaction pattern looks like agent activity. The issuer approves.

Now ask what any of that verified. Not the recommendation. The 4.8-star average the agent trusted is built on review data that the platforms themselves admit is massively polluted. Amazon proactively blocked more than 275 million suspected fake reviews in 2024 alone (Amazon), and the FTC’s Consumer Review Rule, two years old as of October 21, has collected just $750,000 in actual payments across all its enforcement actions. Regulation samples the fraud. Platform filters miss a share of it. And the reviews that survive are increasingly read first by machines.

This is the part that changes qualitatively in the agent era, and it is why the timing of World’s launch matters for a reason Blania did not name. The latest FirstPageSage behavioral dataset found that 61.4 percent of US shoppers now use AI for product research, while only 6.8 percent complete an agent checkout with approval and 1.3 percent go fully autonomous. The hinge of the entire funnel is the approval moment: the human approves what the agent read, without re-reading it themselves. Delegated proof strengthens the approval’s authenticity. It does nothing for the evidence’s.

Why perks for agents raise the value of poisoning the evidence

Here is the uncomfortable corollary of Sada’s generosity thesis, and it deserves to be stated plainly.

Every incentive that makes a verified agent attractive to merchants makes the agent’s information diet attractive to manipulators. Today, fake reviews are a volume game played against scattered human attention: millions of shoppers skim, most distrust what they read, and the conversion rate on any single fake review is low. An agent economy inverts that economics. One successful manipulation of a widely consulted data source, a review corpus, a ratings aggregate, a “best of” list, propagates into thousands of recommendations delivered by trusted agents to humans who have been trained by the approval flow to trust the summary, not the source.

Identity infrastructure makes this worse before it makes it better, in a specific and predictable way. Once merchants can filter for verified agents and offer them perks, agent traffic becomes the preferred channel: higher conversion, lower fraud, cleaner data. The rational spend for anyone selling a bad product shifts accordingly. You no longer need to fool millions of humans. You need to fool the small number of data sources their agents read. Proof of Human guarantees a human authorized the purchase. It is silent on whether a review farm authored the inputs.

This is not an argument against identity infrastructure. The identity layer is necessary, and World’s zero-knowledge approach, which verifies a fact without harvesting the underlying personal data, is the right shape for it. It is an argument that identity is half a trust stack, and the half being built fastest is the half that was already succeeding. Payments got consent. Identity got proofs. The product decision, the moment where money turns into a specific item, still runs on raw marketplace data that the FTC, Amazon’s own fraud teams, and every independent auditor agree is unreliable.

The missing proof: evidence

Zoom meetings now have a Human Badge: a visible, cryptographic signal that a real, unique person is present in a high-stakes conversation. Purchases need the equivalent for products. Call it proof of evidence: a signal, independent of the marketplace selling the item, that the review data behind a recommendation was screened for manipulation and the score reflects review quality rather than review volume.

That is precisely the layer GoBuy builds. Before a purchase is approved, not after:

  • Fake reviews are filtered out and authentic ones weighted up, so the star average an agent cites is built on verified human experience, not review-farm output.
  • The Smart Score, 0 to 100, is computed from review quality, not quantity, because five hundred purchased five-star reviews should score lower than eighty detailed, verified, critical ones.
  • Only the top seven products surface per category, because an agent that returns a thousand options has not made a decision, it has deferred one.
  • The GoBuy Verified badge requires a Smart Score above 80 sustained over 90 days, the same logic as a liveness check: trust is a continuous state, not a one-time event.
  • Everything is exposed to agents over MCP at gobuy.ai/api/mcp, so any shopping agent, including one carrying a World delegated proof, can consult an evidence layer that is structurally independent of the marketplace it is auditing.

The verification stack of October 2026 can now tell a merchant that a human sent this agent. GoBuy tells the human whether what the agent found is worth buying. Both proofs belong in every transaction. Only one of them exists at scale today, and it is the one that was never the binding constraint.

The next time an agent asks for approval on a purchase, ask what it read. If the answer is a star average, the identity stack has done its job and the decision stack has not even started. Check the evidence first at gobuy.ai, or wire the evidence directly into your agent with the GoBuy MCP integration.