AI Shopping Agents Can Now Prove a Real Human Sent Them
The rise of AI shopping agents has made online commerce faster and more convenient than ever. But it has also opened the door to new waves of fraud, bot abuse, and questions about who — or what — is really clicking "buy." World, the identity verification company co-founded by Sam Altman, has launched AgentKit, a new software tool that lets commercial websites verify a real human being is behind every AI-powered purchase. If this works the way it promises, it could reshape trust on the internet.
 |
| Credit: JUAN MABROMATA/AFP via Getty Images / Getty Images under a license. |
What Is AgentKit and Why Does It Matter Right Now
AgentKit is a software development kit released in beta form by Tools for Humanity, the startup that builds and operates the World platform. The tool is designed specifically for agentic commerce — the fast-growing practice where consumers delegate browsing and buying tasks to AI programs that act on their behalf.
The timing could not be more critical. AI agents are no longer a futuristic concept. Millions of people are already using them to search for deals, compare prices, and complete purchases without lifting a finger. This wave of automation is genuinely useful, but it has also created a serious problem: online platforms cannot easily tell whether an action is being taken by a consenting human or an autonomous bot operating without proper oversight.
AgentKit aims to close that gap. It gives website developers a verification layer they can plug into their platforms, so that when an AI agent shows up to make a purchase, the site can confirm a verified human approved that action. For retailers, payment processors, and any business vulnerable to automated fraud, this is a meaningful step forward.
How World ID Powers the Entire System
The backbone of AgentKit is World ID, the identity credential at the center of the World ecosystem. A World ID is a unique, encrypted digital code that represents a verified human being. The most secure version is created through a biometric scan of the user's iris using a physical device called the Orb.
The Orb converts the unique patterns of a person's eye into a cryptographic identifier. This identifier is then stored in the World app and can be used to prove personhood across a growing range of online services — all without revealing sensitive personal data like a name, address, or government ID number. The system is built around privacy by design, which is one of its most important selling points in an era of data breaches and identity theft.
AgentKit plugs World ID into a newly introduced standard for AI agents. When a user instructs their AI shopping assistant to make a purchase, the agent can carry a verified token linked to the user's World ID. The receiving website checks that token, confirms a real human is behind the request, and proceeds with confidence. No human, no transaction — at least in theory.
The Fraud Problem That Made This Tool Necessary
To understand why AgentKit matters, it helps to understand just how bad the bot problem has become. Fraudulent bots already cost the global e-commerce industry tens of billions of dollars each year through fake account creation, inventory hoarding, loyalty point theft, and payment fraud. The mainstream arrival of capable AI agents is expected to make that problem significantly worse.
Unlike the simple bots of a decade ago, modern AI agents are sophisticated enough to mimic human browsing patterns. They can solve basic security challenges, navigate complex checkout flows, and make purchase decisions that look completely legitimate to most fraud detection systems. The old tools built to catch bad bots are struggling to keep pace.
The issue is also not just about bad actors. Even well-intentioned AI agents create problems when websites cannot distinguish them from malicious automation. Retailers may throttle or block all agent traffic as a precaution, which hurts legitimate users who want the convenience of delegation. AgentKit creates a path where good agents can prove their legitimacy and get through, while bad ones cannot.
Sam Altman, OpenAI, and the Irony No One Is Ignoring
There is an uncomfortable irony sitting at the center of this story. Sam Altman co-founded World with the explicit goal of creating proof-of-human technology for an internet increasingly polluted by AI-generated content and automated activity. His other company, OpenAI, is widely credited — or blamed, depending on your perspective — with accelerating that same pollution by making powerful AI tools broadly accessible.
Critics have pointed out the contradiction plainly. The tools that made AI-generated slop, spam, and bots mainstream were largely pioneered by the same ecosystem Altman helped build. Now, World is selling the solution to a problem that AI has significantly amplified.
Supporters of the project offer a different read: that Altman saw the verification problem coming early and started building the infrastructure to address it before most people recognized the threat. Whether that framing is generous or accurate is a matter of debate. What is not debatable is that the demand for human verification tools is exploding, and World is positioned at the front of that market.
What Agentic Commerce Looks Like in Practice
For most people, the concept of AI shopping agents is still abstract. But the reality is closer than it appears. Several major AI assistants already have the ability to browse retail websites, read product listings, compare prices, and initiate purchases on behalf of users. Early adopters are using these tools to handle grocery orders, find the cheapest flights, and buy tickets before they sell out.
As these capabilities mature, the expectation is that a significant portion of online retail traffic will come from agents rather than humans typing directly into browsers. Some projections suggest that within a few years, AI-driven transactions could account for a substantial share of e-commerce volume. Brands and platforms that are not prepared for this shift risk being overwhelmed by traffic they cannot authenticate or monetize properly.
AgentKit is trying to get ahead of that shift. By creating a standard that links agent actions back to verified humans, it gives the entire commerce ecosystem a shared language for trust. A retailer using AgentKit knows not just that an agent is placing an order, but that a specific, verified human being authorized it.
Privacy, Biometrics, and the Questions Worth Asking
No discussion of iris scanning and biometric identity is complete without a serious look at the privacy implications. World has faced substantial scrutiny since its launch, with regulators in several countries raising concerns about how biometric data is collected, stored, and potentially misused.
The company maintains that the Orb does not store raw iris images. Instead, it converts the scan into a mathematical code, deletes the original image, and only retains the encrypted identifier. Users are told they retain control over their data and can delete their World ID if they choose.
Those assurances have not satisfied everyone. Privacy advocates argue that any system requiring biometric enrollment creates risks that go beyond the company's current intentions. If the World ID database were ever breached or misused, the consequences would be uniquely serious — unlike a password, you cannot change your iris. These are legitimate concerns that deserve ongoing scrutiny as the technology scales.
Rebuilding Trust on the Internet
World's AgentKit is not just a product launch. It is a bet on a particular vision of the internet's future — one where proof of personhood becomes as fundamental as an email address or a phone number. If that vision takes hold, it would represent one of the most significant shifts in how online identity works since the web was created.
The challenge is adoption. For AgentKit to fulfill its purpose, it needs to be integrated by enough websites to matter. It needs enough users to create World IDs to make the verification meaningful. And it needs to earn trust in a market that is rightly skeptical of any centralized system claiming to solve the identity problem once and for all.
Those are enormous hurdles. But the urgency is real. The internet is increasingly a place where automated agents outnumber human users in many categories of traffic. Finding a credible, privacy-respecting way to distinguish the two is not a luxury. It is fast becoming a necessity.
AgentKit is one answer to that challenge. It may not be the last word, but for now, it is one of the most concrete attempts to bring human accountability back into the heart of AI-powered commerce.
