Tiago Sada explains why AI is breaking online trust, how World ID tries to prove unique humanhood without exposing identity, and why the privacy objections are the whole story.
This is one of the coolest interviews we’ve ever done.
That sounds like hype, especially for a conversation about identity infrastructure, cryptography, and a shiny orb that scans your face and eyes. But this one works because it starts with a problem almost everyone already understands.
The internet has a human problem.
Bots can solve CAPTCHAs. AI can generate fake selfies and passports. Deepfakes can show up on video calls. Fake accounts can swarm dating apps, games, ticketing systems, social platforms, banks, and comment sections before a normal person can even find the checkout button.
So Corey sat down with Tiago Sada, Chief Product Officer at Tools for Humanity, the company building the technology behind World and World ID, to ask the obvious question:
How do you prove there is a real human on the other side of the screen?
Tiago’s answer is weird, ambitious, and much more practical than the headline version sounds. World wants to create something like a human passport for the internet, but with a crucial constraint: the passport lives on your phone, apps do not learn who you are, and World says it avoids storing the personal images used to verify you.
Now, we know what you're thinking:
"A private company proving you are human? A biometric face-scanning device? A trust layer that could sit underneath dating apps, games, banks, workplaces, video calls, agents, and government credentials that could doxx me? What, is the public internet as we know it over?" I think the next line you'd say after that rhymes with "bell to the gnaw."
And that is what makes the interview worth watching. Corey does not treat proof of human as an easy answer. He pushes directly into privacy, biometrics, legal identity, anonymity, agent delegation, and the possibility that the internet may need a new trust layer without giving any one actor too much power.
Plus, this also feels bigger than identity to us. Once agents can prove they represent a real human, they can also start negotiating access, permissions, and payments across the web, which is where standards like x402 start to look a lot more important. We'll get into that a bit below as well.
Corey and Tiago cover the whole proof-of-human stack in the full episode, from why CAPTCHAs are failing to how World ID uses zero-knowledge proofs and anonymized multi-party computation.
If you want to jump around, start here:
Below, we dive into the rest much more in depth.
Tiago starts with the basic failure mode: the web’s human-detection systems were built for dumber bots.
A CAPTCHA is basically a little intelligence test. If you can solve the puzzle, the website assumes you are human. If you cannot, the website assumes you are a bot. That made sense when bots were bad at visual reasoning, language, and browsing behavior.
AI flipped that assumption.
At (3:24), Tiago walks through the list of old defenses: CAPTCHAs, phone numbers, email accounts, cookies, device fingerprints, payment cards, and document verification. Each one carries an assumption that now looks weaker.
Document checks are the clearest example. A person uploads a passport photo and a selfie, and the system treats that as meaningful proof. AI can now generate fake documents, fake faces, and fake selfie-style verification materials at a quality level that pushes every detector into an arms race.
That arms race matters because the defender has to catch the next fake. The attacker only has to pass once.
Tiago’s point at (16:57) is that “AI detecting AI” is structurally brittle. A detection model catches what it has seen before. The next generation of synthetic media can train against the detector itself.
The ticketing example makes the problem concrete.
At (5:03), Tiago uses the example of a Taylor Swift ticket sale. If Ticketmaster wants every person to get four tickets, the analog-world solution is simple: make people walk to a ticket counter and stamp their hand.
Digital systems lost that stamp.
That is the core World ID pitch: bring back the hand stamp, but make it reusable online.
The best part of Corey’s interview comes early, at (7:35), when he separates three ideas that most people collapse into one phrase.
Those sound similar until you test them against real examples.
For a video game, proving that a player is human may be enough if the goal is blocking a bot from playing the match.
For a social network, uniqueness matters more. A spammer can be a real human and still run 1,000 accounts. The platform needs to know whether one person is creating an army.
For a payment, deployment, or contract signature, authorization matters most. An AI agent might do 99% of the work, but a human should approve the moment where money moves, infrastructure changes, or a legal document gets signed.
That is why Face ID alone does not solve the full problem.
A phone can know a person is currently looking at the device. It has a much harder time proving this is a unique person across the broader internet and that the same person only claimed one allocation, verified one account, or approved one sensitive action.
World ID’s pitch is that those three claims should be separable.
An app should be able to ask for the narrow proof it needs.
A ticketing system may need uniqueness.
A Zoom call may need human continuity.
A website interacting with your AI agent may need proof that a human stands behind the agent.
A government service may need a separate credential, like eligibility or age, issued by the relevant institution.
That is the frame that makes the whole interview click: World ID is trying to separate humanhood from identity.
Tiago is clear at (11:12): World is not anti-bot.
The internet is moving toward agents whether platforms like it or not. Some bots scrape, spam, cheat, and attack. Other bots will soon book your travel, order your food, schedule your meetings, compare products, run research, or operate business workflows.
Websites have historically treated bots as hostile because many bots were hostile. A restaurant delivery app, ticketing platform, online store, or cloud API may block automation because automated traffic often means scraping, fraud, or abuse.
Then a user asks an AI agent to buy a burger.
At (13:28), Tiago walks through that dilemma. A service like DoorDash may want to block bad bots, but it may also want to let a verified user’s agent complete a real purchase.
The missing layer is delegation.
At (14:18), Tiago describes something like digital power of attorney. Your agent could show up to a website and say: yes, I am a bot, but I am acting on behalf of this verified human.
That has two important consequences.
First, useful agents get a way through the door.
Second, abusive agents become accountable. If an agent misbehaves, the platform can block future agents delegated by that human, instead of playing whack-a-mole with infinite bot instances.
The higher-stakes version is “human in the loop,” which Tiago explains at (15:06). An agent can do the work, but when it reaches a sensitive moment, like sending a wire transfer or deploying a production database, it must ask the verified human to approve the action.
That idea is already showing up in World’s product announcements. World has described “human in the loop” integrations for agentic workflows, where a developer can require a zero-knowledge proof that a real, unique human authorized a specific action. World has also announced agent-related work with companies including Vercel, Okta, Browserbase, and Exa.
This is where the proof-of-human question becomes much bigger than login security.
The agentic web needs a way to tell the difference between automation that serves a real person and automation that floods the system.
Yes, we know what half of you are probably thinking.
“Wait, a private company wants to prove I’m human?”
“You think I would actually trust anyone with my biometric data?”
“Does this mean giving up anonymity online?”
“Are we really building a central source of humanhood now?”
I mean... Same.
Those are the right questions.
That is why this interview is important. Corey presses into those exact concerns, especially around (33:10), when he asks what World actually stores, and around (41:16), when he asks to clarify whether World ID proves “I am Corey Noles” or simply “I am a human.”
Thankfully, Tiago gives one of the clearest explanations we have heard of what World ID is trying to do, what it does NOT claim to do, and how the technical design actually works.
The biggest mistake people make with World ID is treating it like a crypto story.
Back in 2021-2022, that was kinda the narrative. Prove you're human, and you'll receive World Coins.
In actuality, it is trust infrastructure: a way for apps to answer three questions without turning every website into an airport security line.
At (7:35), Corey separates the three questions that matter:
That shows up differently depending on the app.
On Tinder, Tiago says verified humans can get a trust boost because the platform knows there is a real person behind the profile. On Zoom, the interesting use case is deepfake protection: proving the face on the call matches the verified person behind the camera. In Shopify stores, merchants could give discounts or priority to verified humans, or to verified agents acting for humans.
Tiago runs through several of those app examples around (22:28). Then he gets into the Zoom face-match example around (42:43), where World ID can help prove that the person on the video call is the same verified person behind the camera.
With Concert Kit, the pitch gets painfully easy to understand: stop bots from grabbing all the concert tickets before the actual fans do and jacking up the price.
Tiago explains the ticketing use case around (24:38), then describes the San Francisco human-only concert stress test around (27:35). Free tickets are basically bot catnip. If you can automate claims, you can hoard access instantly. The point of Concert Kit is to give artists a way to reserve tickets for verified humans without ripping out the ticketing systems they already use.
The agent angle may be the biggest one.
Tiago’s example was a food-ordering agent trying to use a service like DoorDash. Today, many platforms block bots because bots usually mean scraping, fraud, or abuse. Soon, some bots will be useful agents trying to buy lunch for a real person.
That creates a new category: “yes, I am a bot, but I am acting for this verified human.”
Tiago calls that agentic delegation around (14:18). It is basically digital power of attorney for the agent era. Your agent can do useful work on your behalf, and the website gets a trust signal that the action is tied to a real, unique human.
Personally, we see another important use case here: combine proof of human with x402, and it could become easier for publishers to get paid when bots view their content.
Think of it like micro-payments using credits or tokens for pay-per-view content. If I send my bot to your website, it can read the article without me seeing an ad, and you still get paid.
We already pay agents per token. Why not pay for content the same way?
Don't get me wrong, I'm not advocating for the end of the free ad-supported internet.
Using this model, free content can still thrive. But this model could empower paid content to become more accessible to more people in an agent-friendly future. If agents can authenticate, pay, and access information without forcing every user into a subscription funnel, that means 1. more people can read more important news from high quality sources, and 2. those publishers can get rewarded for publishing more high quality content.
Put simply, it could help keep publishing alive on the internet without banning all the bots. In a world increasingly rebuilding itself around agents, it could divert publishers away from the painful path of suing their way to partial, temporal victory in the legal system and instead create incentives for more people to keep making high-quality work. Robots.txt would get replaced with Robots.pay-me-play. Or to paraphrase Rihanna... "bot better have my money!"
And that is the deeper point: the AI internet needs more than better bot detection.
Fundamentally, it is much easier to verify humans than identify every possible bot. AI systems will keep changing. Bad bots will keep adapting. Good agents will increasingly look like bad bots unless there is a way to prove they are acting for a real person.
So humans, agents, and platforms need a way to negotiate trust without exposing everyone’s identity everywhere.
This kinda seems like the answer, y’all.
Let’s start to embrace it, while staying sharp about the governance questions that come with it.
However, before we embrace it, we must understand how it actually works...
The simplified version is this: World ID tries to prove you are a real, unique human without making every app learn who you are.
The actual flow has several layers.
At (19:54), Tiago describes World ID as a “blue check mark for the internet that actually works.” You download the World App, and a passport-like credential is created on your phone.
Some apps may accept lightweight verification. Tiago says at (20:13) that users can take a selfie for a lighter check, use NFC-enabled IDs in some countries for a medium level, or visit an Orb for the highest level.
At (21:47), Tiago walks through the Orb flow. You pair your phone with the Orb by scanning a QR code. The device looks at you for about 15 seconds and checks that you are a real person, rather than a mask, a screen, a dog, or another attempt to cheat the system.
The point is uniqueness at population scale. A website can already ask whether a person is in front of one device. World is trying to determine whether that human has already verified somewhere else.
This is the first major privacy claim. Tiago says at (21:59) that the images taken by the Orb get sent to your device and then deleted from the Orb.
World’s public materials describe the same basic architecture: after Orb verification, data is encrypted, sent to the phone, and permanently deleted from the Orb. World also describes “Personal Custody” as a system where information generated at the Orb is held on the user’s device.
At (35:16), Tiago gets into anonymized multi-party computation, or AMPC.
The plain-English version: your verification data gets transformed, scrambled, and split into fragments. Those fragments are distributed to trusted universities so no single party has a meaningful version of your underlying biometric data.
Tiago’s analogy at (36:40) is paint. Imagine you put yellow, green, red, and blue on a palette. Then you mix them together, cut the result into pieces, and hand those pieces to different people. One person holding one scrambled piece cannot reconstruct the original palette.
Then, the system can still perform the narrow calculation it needs: has this person verified before?
That is the heart of the uniqueness problem. The system needs to catch duplicates without storing the sort of central biometric database that would make everyone nervous.
World’s technical approach also lines up with broader work on secure multi-party computation. One paper on private iris-code uniqueness checks describes protocols designed to query whether an iris code is already in a database while protecting both the query and the dataset. The big idea is the same: run the comparison without exposing the raw sensitive data.
At (38:18), Tiago explains the role of zero-knowledge proofs.
A zero-knowledge proof lets you prove something without revealing the underlying information.
In this case, the proof might say:
The app gets the proof it needs, rather than the user’s biometric data, name, birthday, address, or global identity.
World’s public FAQ says the same thing: when a person uses World ID, zero-knowledge proofs prove that the person is a unique human without sharing personal information with the third-party service.
At (39:02), Tiago explains the disposable-ID idea.
When you use World ID with one app, that app sees a generated version of your credential. Another app sees a different one. The goal is to prevent apps from comparing notes across your life.
Tinder should not see the same identifier as a ticketing platform.
A Shopify store should not see the same identifier as a game.
A workplace system should not see the same identifier as a social app.
Tiago’s claim is stronger: World itself cannot see all the different disposable IDs you have generated across apps. Only the user’s phone knows that full map.
This is the part that makes the system conceptually different from a universal login. A universal login follows you around. World ID’s pitch is that it can prove one fact at a time without becoming a single tracking handle.
The most important identity distinction comes at (41:16).
Corey asks, when using World, whether a user is verifying “I am a human” or “I am Corey Noles, the human.”
Tiago’s answer is basically: by default, World ID proves the first thing.
The Orb verification proves that you are a real, unique human. It does not automatically prove your legal name, your passport identity, your address, your nationality, or your age.
That is a feature of the design.
Legal identity is a separate credential layer. Tiago says governments should be the ones issuing legal identity credentials. But he also says governments and institutions may eventually issue credentials using the same rails.
At (45:38), he gives the age-verification example. If a government issues your driver’s license credential in a compatible format, a website could ask you to prove you are over 21 without receiving your full driver’s license.
That changes the mental model.
A traditional identity check asks for too much. A site asks for a date of birth and receives name, address, ID number, issuing authority, photo, and other sensitive details. I seriously hate filling all that out. Then, the site has to store or process data it probably never wanted in the first place.
A credential-based proof asks for one fact.
At (47:01), Tiago describes World ID as an identity protocol underpinned by proof of human. The base credential is humanhood. Other credentials could be issued by governments, companies, universities, or employers.
That is the subtle but powerful idea: World ID is less interesting as “one company judges all humans” and more interesting as a privacy-preserving credential standard where different institutions can issue different proofs.
The base layer says: real, unique human.
The next layers can say: over 21, enrolled, employed, authorized, eligible.
The app only gets the claim it needs.
Tiago makes this much easier to understand by walking through consumer examples.
At (22:28), he mentions Zoom, Tinder, Vercel, Exa, Browse, and Shopify-style use cases.
Zoom is about deepfake protection. World describes a three-way match: the cryptographically signed image from Orb verification, a real-time Face Auth selfie on the user’s device, and the live video frame that other participants see. The goal is to prove the person on the call is the expected verified human.
Tinder is about profile trust. World says Orb-verified users can display a verified human badge in select markets, with promotional boosts attached.
Razer and gaming integrations are about keeping bot armies out of places where human competition matters.
Concert Kit is the most emotionally obvious example. Other Swifties know what I'm talking about at a deep emotional level, and where this is going next...
At (24:22), Tiago describes World’s partner categories: gaming, dating, and ticketing. Then at (26:54), he explains Concert Kit as a way for artists to reserve a portion of tickets for verified humans, similar to fan-club presales.
World’s Concert Kit announcement frames the problem bluntly: bots can buy thousands of tickets in seconds, push them to resale markets, and leave real fans paying double, triple, or worse. World also cites examples like the Taylor Swift Eras Tour presale, where Ticketmaster reported billions of system requests in a single day.
At (27:35), Tiago says Thirty Seconds to Mars is using World ID to reserve a portion of tickets for verified humans on an upcoming tour. He also describes a San Francisco event with Anderson .Paak performing as DJ Pee .Wee and St. Vincent performing as DJ St. Vicious.
The fun detail comes at (28:19): Tiago says hundreds of thousands of bots attacked the site trying to claim free tickets. The venue capacity was about 2,000, and roughly 1,000 people claimed tickets, with each verified human able to claim up to two.
That is a good stress test because free tickets are bot bait. If tickets cost nothing, any automated advantage becomes even more valuable.
The simplest version of the World pitch is right there: let the humans in and keep the bots outside.
The strongest critique of World ID is that technical privacy and social trust are different problems.
A system can use elegant cryptography and still raise real concerns about power, incentives, access, coercion, and mission creep.
Will people feel free to opt out if the best apps, ticket sales, games, and enterprise tools increasingly reward verified humans?
Will the benefits flow evenly across countries, devices, incomes, and accessibility needs?
Will third-party apps respect the narrow-proof model, or will they gradually ask for more credentials because more data is always tempting?
Will governments treat privacy-preserving credential rails as a better alternative to ID uploads, or as an easier way to make proof requirements more common?
World’s answer is that the protocol should be open source, privacy-preserving, and optional. Tiago says at (40:30) that the hardware, firmware, AMPC system, zero-knowledge proofs, and SDKs are open source because the company wants others to inspect, audit, and reuse the primitives.
That helps.
It does not settle the governance question.
Biometric systems deserve more scrutiny because the stakes are higher. Critics have raised concerns about biometric data, re-identification, consent, and whether users fully understand what they are joining. Regulators have already pushed back in multiple countries.
The interesting question is whether privacy-preserving cryptography, self-custody, open-source primitives, and narrow proofs can give the internet a better option than bot chaos on one side and universal ID checks on the other.
The internet used to assume that most users were people, most images were camera outputs, most documents came from institutions, and most accounts took some effort to create.
AI erodes those assumptions.
That does not mean every website needs proof of human. It means some parts of the internet will need stronger guarantees than “this browser session passed a puzzle.”
The places with scarce resources will feel it first:
The places with high impersonation risk will feel it too:
In each case, the old question was: does this account have the right credential?
The new question is: is the right human actually present, or did automation take over the shell of trust?
That is what World calls human continuity. The system is trying to verify the human behind an interaction, rather than merely the device, password, account, or token.
For everyday users, the best version of this future feels simple. You verify once, carry proof on your phone, and reveal only the claim required in the moment.
For developers and platforms, the best version creates a new building block. You can add a human gate at moments where abuse, fraud, or agent overreach would otherwise break the system.
For governments and institutions, the best version moves identity checks away from oversharing. Instead of uploading your full ID everywhere, you prove one narrow credential.
For agents, the best version creates accountability. Helpful agents can act. Abusive agents have a human sponsor that platforms can throttle or block.
That is a big shift.
It also leaves one unresolved issue that will decide whether proof of human becomes trusted infrastructure or a new source of friction: who gets to set the rules for when human proof is required?
If the answer is “every app asks for everything” or you don't get access, this becomes a privacy failure.
If the answer is “apps ask for the minimum proof needed,” this could be a better trust layer for an AI-filled internet.
We think Tiago’s interview is worth watching because it treats that tension seriously. The internet needs a way to tell humans, bots, and human-backed agents apart. The acceptable version has to preserve anonymity, avoid central biometric databases, separate legal identity from humanhood, and let users prove one fact at a time.
That is the whole bet.
World ID is not only trying to prove you are human.
It is trying to prove the internet can know that without knowing everything else.
By God, I hope it works. For all of our sakes.
Grant Harvey is the Lead Writer of The Neuron, where he continues to lead the publication's daily coverage of AI news, tools, and trends.
Don't fall behind on AI. Get the AI trends & tools you need to know. Join 700,000+ professionals from top companies like Microsoft, Apple, Salesforce and more.
Property of TechnologyAdvice. © 2026 TechnologyAdvice. All Rights Reserved
Advertiser Disclosure: Some of the products that appear on this site are from companies from which TechnologyAdvice receives compensation. This compensation may impact how and where products appear on this site including, for example, the order in which they appear. TechnologyAdvice does not include all companies or all types of products available in the marketplace.
