Proof of Human Is Not KYC: Crypto Biometrics and AML in 2026

On 18 October 2026, the identity company co-founded by Sam Altman did something that would have sounded absurd three years ago: it tried to sell a bank an eyeball. World, the project built around a chrome sphere that scans your iris, launched a suite of proof-of-human tools aimed squarely at financial institutions, pitching biometric verification as the missing trust layer for an internet where you can no longer assume the face on the video call, or the selfie in the onboarding flow, belongs to a real person. The release leaned on a Deloitte forecast that AI-enabled fraud losses in the United States could reach $40 billion by 2027, up from an estimated $12.3 billion in 2023.

The timing is not subtle. Crypto spent 2026 learning that identity, not price, is the industry’s softest surface. Bitcoin changing hands around $82,000 on the day of the launch (CoinGecko put it near $81,900, a market capitalization of roughly $1.66 trillion, per Fortune’s daily quote) barely moved on the news. What moved was the question underneath it: if a deepfake can pass a crypto exchange’s identity check for the price of a streaming subscription, can an iris scan save the whole know-your-customer regime?

This piece argues that the answer is no, at least not in the way the pitch implies, and the reason is a distinction the marketing papers over. Proving that you are a unique, living human is a genuinely hard and genuinely useful thing. It is also not the thing that anti-money-laundering law actually requires. The law wants a name.

KYC, AML, and who actually enforces them

Start with the vocabulary, because the two terms get mashed together and they are not the same. Anti-money laundering (AML) is the whole legal regime that obliges regulated firms to detect and report the movement of criminal money. Know your customer (KYC) is one piece of it: the set of checks a firm runs to establish who its customer is, before and during the relationship. KYC is the front door; AML is the house.

In the United States the rules flow from the Bank Secrecy Act and its amendments, and the agency that writes and enforces them is the Financial Crimes Enforcement Network (FinCEN), a bureau of the Treasury, working alongside the Office of Foreign Assets Control (OFAC) on sanctions. This trips up newcomers constantly, so it is worth saying plainly: the Securities and Exchange Commission does not run crypto AML. The SEC polices whether a token is a security; FinCEN and OFAC decide whether an exchange is laundering money. A crypto business that touches customer funds is usually a money services business, and a money services business owes the full program.

That program rests on five pillars: a designated compliance officer, written policies, employee training, independent testing, and risk-based customer due diligence. The laundering it is built to catch runs in three classic stages: placement (getting dirty cash into the system), layering (obscuring its trail through many hops), and integration (bringing it back out looking clean). KYC is the control that is supposed to stop stage one at the door, by attaching a verified identity to every account. Hold that phrase, verified identity, because it is where the biometric story runs into trouble.

What the law asks for: a name, not a face

The sharpest way to see the gap is to read the actual rule. A US Customer Identification Program, mandated by Section 326 of the USA PATRIOT Act and codified at 31 CFR 1020.220, requires a firm to collect, at a minimum, four things from an individual before opening an account: full legal name, date of birth, residential address, and an identification number (a Social Security or taxpayer number for US persons, a passport or similar for others). The firm must then verify that information and, in the regulation’s words, form a “reasonable belief that it knows the true identity of each customer.” It has to keep records of how it did so, and it has to check the customer against government lists.

That last step is the one biometrics cannot touch. OFAC sanctions compliance means taking a customer’s identifying details and screening them against the Specially Designated Nationals list, and every other list that applies. You cannot screen a person you cannot name. A proof that someone is a unique, breathing human returns exactly zero of the data points the screen needs. The sanctioned oligarch, the fentanyl broker, and the North Korean IT worker are all unique living humans too.

Read that table and the design problem is obvious. Everything KYC is legally for sits in rows the biometric cannot fill. Everything the biometric is good at sits in a single row the statute does not actually ask about. That does not make proof of personhood worthless. It makes it a different product wearing the KYC label.

How exchanges check you today: document, selfie, liveness

To understand why biometrics crept into KYC at all, look at how a modern exchange onboards you. You photograph a government ID, then take a selfie, and software compares the two faces while a liveness check tries to confirm the selfie is a live person rather than a photo of a photo. Vendors such as Sumsub, Jumio, Onfido, Persona, and iProov sell this stack, and it quietly became the default because it is cheap, fast, and remote. The biometric here is not the iris; it is the face match plus the liveness test, and it carries real regulatory weight, because it is how a firm bridges from the name on the document to the human in front of the camera.

Crucially, this flow still begins with a document that carries a name. The face match exists to bind the human to the name, not to replace it. That is the correct architecture for AML, and it is also, as the next section shows, the layer that has collapsed.

The layer that broke

Face-plus-liveness worked until generative AI learned to fake both halves. The 2026 threat picture is stark. iProov, one of the liveness vendors, reported that injection attacks against its iOS pipeline rose 1,151% in the second half of 2025, a 741% jump year on year, as attackers piped synthetic video straight into the camera feed. Sumsub’s annual report found that synthetic identity document fraud surged more than 300% in the United States, and that synthetic identities, invented people stitched from real and fake data, are now the single largest first-party fraud method at 21% of attempts.

The tools are commodities. Real-time face-swap, camera injection, and voice cloning are sold as fraud-as-a-service for less than $50 a month, and a competent operator needs a laptop and about half an hour to beat a standard check. One underground service, OnlyFake, advertised AI-generated photo IDs for roughly $15 and reportedly passed the onboarding check at a major exchange. The crypto sector takes the worst of it: it has carried the highest fraud rate of any industry Sumsub tracks, and in one reported incident an AI-generated handheld video impersonated a MEXC account holder well enough to move around $340,000 before anyone caught it.

So the industry has a genuine, expensive wound, and proof of personhood is marketed as the dressing. The logic is seductive: if you cannot trust that a face is live and unique, enroll each human once, at a trusted device, and hand them a credential no bot can clone. That is a real fix for a real problem. It is just a fix for the wrong problem, if the problem you are being sold on is AML.

Proof of personhood, explained

Proof of personhood is the attempt to give every real human exactly one digital token of uniqueness, without necessarily knowing who they are. The idea predates World. Ethereum co-founder Vitalik Buterin has written about it since at least 2023, framing it as the answer to Sybil attacks, the trick of one actor spinning up thousands of fake identities to capture a vote, an airdrop, or a market. If you can prove one-person-one-account, a whole class of manipulation dies. It is the same weakness that lets bots quietly set the odds in prediction markets and lets wash traders fake volume.

World’s mechanism is the Orb, a polished sphere that photographs your irises, converts the pattern into a numeric code, and, per the project’s own description, deletes the original image while the code is split into fragments across separate servers. When an app asks whether you are a verified unique human, the World App produces a zero-knowledge proof that your code sits in the global set of verified codes, without revealing which code is yours. The cryptography is the same family that powers zero-knowledge proving markets elsewhere in crypto. By design, World does not ask for your name, email, or government document to issue the basic credential. That anonymity is the whole selling point, and it is also precisely why the credential cannot do KYC.

Inside World’s October pitch to the banks

The 18 October launch packaged three distinct credentials for financial institutions. Proof of Human is the one-time Orb enrollment, attesting that an account belongs to a unique real person. Proof of Presence is a smartphone liveness check, attesting that whoever is on a call or in a session right now is live and not a deepfake; it ships inside World ID for Zoom, which displays a human badge on verified participants. Proof of Attributes verifies a specific fact drawn from a government credential, such as age or nationality, and discloses only that fact. A fourth piece, World ID for Agents, lets a person delegate a proof to an AI agent so a bank can tell an authorized bot from a rogue one.

More than 18 million people already hold a Proof of Human, and the launch named OpenExchange, Bain Capital, VanEck Funds, and Outtake among its design partners. Sandro Herbig, president of Tools for Humanity, framed the product this way: “As AI becomes increasingly prevalent in all aspects of our lives, enterprises are requesting individuals provide ‘Proofs’ about themselves to help mitigate fraud and increase trust. World ID for banking and finance offers a suite of tools to add an AI-safe trust layer across their internal and external systems.” World’s chief executive and co-founder Alex Blania has put the ambition more bluntly, telling reporters that “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.”

Uniqueness is not identity

Here is the crux. A compliance officer at a regulated exchange has to answer two questions about every customer: is this one real person, and who is that person? Proof of Human answers the first with more rigor than any selfie check ever could. It says nothing at all about the second. And the second is the one the statute, the examiners, and the sanctions screen all care about.

Map it against the three stages of laundering and the mismatch is clean. Proof of personhood frustrates the launderer who wants to open ten thousand mule accounts, because each human can hold only one credential, and that is a genuine gain against Sybil-style structuring. It does nothing to stop a single named individual from being a sanctioned party, a politically exposed person who needs enhanced due diligence, or a straw buyer moving someone else’s money, because none of those facts are visible in a credential built to reveal nothing but uniqueness. You can be the most verifiably human customer on the platform and still be exactly the person OFAC wants frozen.

This is why the honest framing is complement, not substitute. Proof of Presence genuinely hardens the liveness layer that deepfakes shattered, and World ID for Agents addresses a problem that barely existed a year ago, namely autonomous software moving money and needing to prove a human stands behind it. Bolted onto a KYC flow that still collects a name and screens it, a strong liveness-and-uniqueness layer is a useful upgrade. Sold as the flow itself, it quietly deletes the part the law was written for. Buterin himself has cautioned that biometric uniqueness is powerful but not self-sufficient, warning of coercion (someone forced to scan an iris against a key that is not theirs), even 3D-printed fake enrollments, and arguing that no single personhood method should stand alone.

Proof of Attributes: where the anonymity quietly ends

There is a tell in the product line. The moment World added Proof of Attributes, pulling age or nationality from a government credential, it conceded the point. A pure proof of human cannot satisfy KYC, so the banking suite bundles a second credential that reaches back to the exact document, the passport or national ID, that proof of personhood was supposed to make unnecessary. Selective disclosure is clever engineering: the bank learns the customer is over 18, or a citizen of a given country, without seeing the full document. But the trust still originates in a state-issued identity, which means the anonymity that made iris-scanning feel acceptable to users is not the thing doing the compliance work.

Follow that thread and proof of personhood stops being an alternative to identity infrastructure and becomes a front end for it. That may well be where the whole category is heading, and it is not necessarily a bad outcome; a privacy-preserving wrapper over verified attributes is a better deal for users than handing a photocopied passport to every exchange and hoping the platform never quietly offboards them. But it is a very different claim from the one on the tin. The eyeball is not doing KYC. The passport behind it is.

The honeypot problem, now with your iris

Every KYC regime builds a honeypot. To verify customers, firms accumulate exactly the dossier a criminal most wants: names, addresses, dates of birth, ID scans, and now faces. In May 2025 that abstraction turned concrete, when Coinbase disclosed that bribed overseas support agents had stolen the personal data of 69,461 customers, including government-ID images; the company refused a $20 million ransom, offered a matching $20 million bounty, and pegged the cleanup at $180 million to $400 million. The lesson stuck with the industry, which has since argued over how to pay for finding these holes before thieves do.

Biometrics raise the stakes of that honeypot in a way passwords never could. You can reset a stolen password; you cannot reissue an iris. World’s answer is real engineering: the raw image is deleted, the derived code is fragmented across servers, and apps see only a zero-knowledge proof rather than the biometric itself. That genuinely reduces the blast radius compared with a vendor sitting on a database of face scans. But it also concentrates a new category of data, irreversible biometric identifiers for tens of millions of people, inside one private network whose deletion claims, as regulators keep pointing out, nobody outside can independently verify. Trust is doing a great deal of work in that sentence.

What the biometric-privacy laws already say

Biometric KYC does not land on a blank legal page. In the United States there is no federal biometric-privacy statute, but Illinois’s Biometric Information Privacy Act (BIPA) has teeth: it requires informed written consent and a published retention policy before a company collects a face or iris scan, and it gives individuals a private right of action worth $1,000 for a negligent violation and $5,000 for an intentional one. A 2024 amendment softened the math by treating repeated scans as a single violation, and in April 2026 the Seventh Circuit held that the damages change applies retroactively. The consent and retention duties remain, and Texas and Washington have their own versions.

Europe is stricter still. Under Article 9 of the GDPR, biometric data processed to uniquely identify a person is a special category whose processing is prohibited unless a narrow exception applies, and consent that is a condition of service is rarely treated as freely given. That is why World’s history reads like a map of data-protection enforcement rather than a growth chart.

The pattern, documented in reporting on World’s country-by-country reception, is consistent: regulators are not objecting to the math of proof of personhood, they are objecting to a private company building a global registry of irreversible identifiers. A firm that adopts biometric KYC inherits that fight, not just the vendor’s uptime.

Does the AML rulebook even allow it?

Suppose a firm wants to lean on a digital or biometric credential anyway. Is that even permitted? Broadly yes, with conditions. The Financial Action Task Force, the global standard-setter whose recommendations most countries translate into law, said in its 2020 Guidance on Digital Identity that a reliable, independent digital identity system can be used for customer due diligence, and that non-face-to-face onboarding can be standard, or even lower risk, when the controls are strong enough. The door is open. The catch sits in the adjectives: reliable and independent, demonstrably so, with the regulated firm keeping the liability if the credential fails.

And FATF is openly nervous about the biometric layer in particular. Its December 2025 horizon scan on AI and deepfakes flagged three structural risks at once: that growing reliance on facial biometrics enlarges the attack surface, that synthetic audio, video, and images defeat KYC and liveness checks, and that cross-border deepfake fraud exploits the gaps between national rulebooks. FATF’s recommended answer is not to trust one biometric; it is layered verification with human review. No regulator has blessed a stand-alone proof of human as sufficient customer due diligence, and nothing in the 18 October launch changes that. World can supply a very good control. It cannot supply the firm’s legal conclusion that it knows who the customer is.

The rules are real, and the fines are larger than the pitch

It is tempting to treat all of this as theoretical. The enforcement record says otherwise. In 2023 Binance admitted to AML and sanctions failures and paid more than $4.3 billion; prosecutors found it had processed over 100,000 suspicious transactions and filed not a single suspicious activity report. Attorney General Merrick Garland’s line at the time became the sector’s epitaph: “Using new technology to break the law does not make you a disruptor. It makes you a criminal.” OKX pleaded guilty and paid more than $504 million in early 2025; KuCoin more than $297 million; BitMEX, the template offshore case, $100 million. The through line in every one was identity: accounts opened with too little verified information, monitored too loosely.

The rulebook is tightening, not loosening. In the United States the GENIUS Act drags stablecoin issuers into the Bank Secrecy Act, with a FinCEN and OFAC rulemaking due to bite in January 2027. In the European Union the Anti-Money Laundering Regulation takes full effect on 10 July 2027 and will ban anonymous crypto accounts and privacy coins outright, policed by a new authority, AMLA, in Frankfurt. Against that backdrop, a credential whose entire pitch is anonymity is swimming upstream in Europe, even as it is marketed as compliance infrastructure.

Whether any of it works is a separate and uncomfortable question. Chainalysis put illicit crypto volume at roughly $154 billion in its 2026 report, still under 1% of all activity, most of it now moving in stablecoins. Critics such as Coin Center’s Peter Van Valkenburgh argue the surveillance apparatus achieves “remarkably little” relative to its cost and the sliver of criminal funds ever recovered. Proof of personhood does not resolve that debate. If anything it sharpens it, by offering firms a shinier control to spend on, while the thing the law actually measures, a name screened against a list, stays exactly as hard as before.

What it means for you

For an ordinary user, the near-term reality is more verification, not less. Expect exchanges and banks to add liveness and proof-of-human layers on top of the document check you already do, because the deepfake threat is real and regulators now expect a response to it. You are unlikely to be allowed to swap your passport for an iris scan, because the law will not let the firm accept that trade.

For an exchange or a builder, the guidance is to buy proof of personhood for what it is: a strong defense against bots, Sybil attacks, and deepfake liveness fraud, and a plausible way to gate AI agents. Do not let a vendor deck imply it discharges your CIP, CDD, or sanctions obligations, because an examiner will not. If you deploy biometrics, treat the data as a special-category liability from day one, with consent, retention limits, and a BIPA and GDPR analysis, rather than an afterthought.

And for the banks World courted on 18 October, the clear-eyed read is that the Orb solves the half of the problem AI created and leaves untouched the half the Bank Secrecy Act was written for. Proof of human is a real, and probably necessary, tool for an internet filling up with convincing fakes. It is just not a passport, a sanctions screen, or a name. KYC still wants the name.

Frequently Asked Questions

Is World ID the same as KYC?

No. World ID is a proof of personhood: it confirms you are a unique, living human without revealing your name, email, or government document. KYC requires a firm to establish and verify who you actually are, collecting your legal name, date of birth, address, and an identification number, then screening that identity against sanctions lists. World ID can strengthen a KYC flow, but it cannot replace it.

Can a biometric scan satisfy anti-money-laundering rules on its own?

Not by itself. US rules at 31 CFR 1020.220 require a Customer Identification Program that names and verifies the customer, and OFAC screening needs identifying details to check against the sanctions list. A biometric proves uniqueness or liveness, not identity, so no regulator currently treats a stand-alone proof of human as sufficient customer due diligence.

Who enforces crypto KYC and AML in the United States?

FinCEN, a Treasury bureau, writes and enforces the Bank Secrecy Act rules, alongside OFAC for sanctions. The SEC does not run crypto AML; it decides whether a token is a security. Most crypto businesses that handle customer funds are money services businesses and owe the full AML program.

Why has World been banned or restricted in so many countries?

Regulators in Spain, Germany, Hong Kong, Brazil, Portugal, Indonesia, the Philippines, and Thailand have halted, fined, or ordered deletion of World’s biometric data, mostly on data-protection grounds. The objection is less about the technology than about a private company building a global registry of irreversible biometric identifiers, often with consent and deletion practices regulators found inadequate.

Does proof of personhood make my data safer or less safe?

It cuts both ways. Systems like World’s delete the raw iris image, fragment the derived code, and share only a zero-knowledge proof, which reduces exposure compared with a database of stored face scans. But biometrics are irreversible, you cannot reissue an iris the way you reset a password, and the data concentrates inside one network whose deletion claims outsiders cannot independently verify.

By Anneke de Vries, regulation desk, HOGE Wire.