World, the web3 project linked to Sam Altman and Alex Blania’s Tools for Humanity, is looking beyond proving that a user is human. Its next possible step is helping websites understand when an AI agent is acting for a real person.
The shift comes as OpenAI’s Operator has brought autonomous web agents into sharper focus. Operator is OpenAI’s first AI agent that can act autonomously on the web, and early use already shows a basic tension: some websites can detect AI agents and block them, while some businesses may want to let trusted agents in.
Why World ID could matter for AI agents
World was formerly known as Worldcoin, and its core idea is that the internet may reach a point where humans and AI agents are difficult to tell apart. To answer that problem, World built “proof of human” tools around World ID.
The process described by the project involves scanning a person’s eyeball with a silver metal orb, then giving that person a unique identifier on the blockchain. That identifier is meant to verify that the person is human.
Tiago Sada, World’s chief product officer, told TechCrunch that the same identity layer could become useful when people rely on AI agents to act for them online. In that model, the important question is not always whether the immediate user is a person. It may be whether a real person has endorsed the action.
“This idea of delegating your ‘proof of personhood’ to an agent and letting it act on your behalf is actually super important,” he said in an interview with TechCrunch. “Instead of only allowing people you think are human [on your website], you will also allow AI agents that represent a real human. This is where World ID comes in.”
That framing turns World ID from a human-verification system into something broader: a way to connect digital actions to a person, even when the action does not come directly from that person’s own account.
From proving people to licensing agents
Sada said World’s ID technology could be used to license AI agents to act on someone’s behalf. A recent World blog post also says its proof of human tools will not only separate humans from bots, as they do today, but could help people control a network of AI agents online.
The project has said its tools for AI will be essential in 2025. Sada also said the company will “need to see” whether any of those tools involve linking AI agents to people.
The direction is still emerging, but the basic use case is clear. A website might not need a human hand on every click. It might need confidence that a human stands behind the agent performing the action.
“There’s certain apps where it doesn’t matter if an actual person is using it, or an agent acting on their behalf. You just care to know there is a person endorsing that interaction,” said Sada.
That distinction matters because AI agents are not simply another kind of account. They may browse, request, purchase, or interact across services on a user’s behalf. If websites cannot tell the difference between a random bot and a user-approved agent, they may default to blocking both.
Websites are already deciding who gets in
Many websites use tools from Cloudflare and Snowflake to stop AI bots from scraping their content. Early users of OpenAI’s Operator found that some websites blocked OpenAI’s new agent by default.
At the same time, businesses may have reasons to reconsider strict blocking. Sada argued that companies ultimately want more users and more sales, and that agents could become a new path to those outcomes.
“At the end of the day, businesses want to sell more. They want to serve more users,” said Sada. “If a delivery app’s sales increase because there’s a bunch of agents coming and replying on behalf of their users, the business is going to be happy about that.”
That example aligns with OpenAI’s collaborations with Uber, Instacart, and DoorDash, which allow Operator to use their platforms. Those companies are used to serving people directly through their apps. Now they are also preparing for user interactions that may be handled by AI agents.
Sada described a future where an agent clearly identifies itself as a bot but also shows that it is acting for a person. His example was direct: “Yes, I’m a bot. Don’t worry, don’t freak out. I’m actually here to buy a hot dog for Tiago,” he said.
The trust problem is also a security problem
Letting every AI agent onto every site would create obvious risks. The source article points to DDOS attacks, scams, and other bad actors as reasons websites cannot simply open their doors to all bots.
World’s proposed role is to create a more selective path. Sada said that by allowing only a couple AI agents per person, businesses could reach users through agents while protecting the integrity of their sites.
That makes the identity layer the central issue. A service does not only need to know that traffic is automated. It needs to know whether the automation is tied to a real person, whether the person endorsed the interaction, and whether the agent should be allowed to proceed.
World’s history adds another layer to the story. Tools for Humanity has changed direction before and has faced controversy. World began as a crypto project, was temporarily banned in some countries, and was recently ordered to let Europeans delete their biometric data on request. In October, the project dropped “coin” from its name, signaling a stronger focus on human verification and less emphasis on crypto.
Sam Altman’s ventures also connect in notable ways. OpenAI is building web agents through Operator. World is working on identity and human verification. Helion Energy, Altman’s nuclear fusion startup, could one day provide power for OpenAI’s data centers. Retro Biosciences, a longevity science startup he backs, is already using OpenAI’s models to try to increase the human lifespan.
The open question is whether World’s tools will become more deeply integrated with OpenAI. Altman speaks at World’s events, supposedly talks with the team every week, and the project continues to lean into AI work. As platforms begin allowing OpenAI’s agents onto their services, World may try to become part of the trust layer that tells websites which AI agents deserve access.