The press release said banks. I read it four times. No bank is named.
What is named: Bain Capital, VanEck, Pantera. Three asset managers. Two are already capital relationships for World. None operates deposit-taking infrastructure, a retail onboarding desk, or the supplier-risk apparatus a bank runs before letting a vendor touch customer identity data.
The headline promised institutional distribution. The body delivered a logo row.
That gap is the story. Not the $40 billion AI fraud figure, which appears with no attribution, no methodology, no source institution. Not the phrase "delegation to AI agents," which is the most technically interesting clause in the release and got exactly one clause. The story is that a company with years of regulatory scar tissue announced entry into the most compliance-sensitive procurement channel on earth and could not produce one regulated counterparty.
The code spoke, but the logic was a lie.
World, formerly Worldcoin, sells three things that get confused for one: an Orb that scans irises, a credential called World ID, and a token called WLD. The credential is the product. The Orb is the supply chain. The token is the subsidy that pays people to stand in front of the Orb.
The pitch: prove you are human, once, carry the proof everywhere.
For four years that pitch had no buyer. Verification was rewarded in WLD, so demand was manufactured by inflation. Then the AI fraud narrative arrived and supplied a marketable enemy.
The release leans on it. Four hundred billion dollars in projected AI fraud losses. That number does framing work, not analytical work. Deepfakes, synthetic identities, and agent impersonation are three problems with three mitigations. A single aggregate collapses them into a marketing asset.
Define the term precisely: "design partner" in enterprise software means a co-design counterparty. Usually unpaid. Sometimes the vendor pays for the logo. It is not a signed customer, not a production deployment, not revenue.
That distinction is the whole announcement.
Here is the causal jump nobody addresses.
Proof-of-personhood solves one problem: at issuance, bind a credential to a unique living human. That is a registration-time assertion. It is not a content-authenticity primitive. It does not detect a deepfake. It does not stop a verified human's account from being socially engineered, or that human from renting out the credential.
If the bank's loss driver is synthetic identity โ real name, fabricated history, manufactured credit file โ then World ID contributes one boolean to a decision requiring dozens of signals. Useful. Not a $40 billion countermeasure.
The clause worth a technical read is the fourth: delegation to AI agents.
If built as a scoped, revocable delegation credential, this is a larger market than bank KYC. Agent payments, agent authorization, agent liability attribution โ each needs an identity layer answering which human stands behind the process. Whoever issues that credential becomes a toll booth on the agent economy.
Delegation is also where these systems break, and they break in a predictable place.
struct Delegation {
address principal; // World ID holder
address agent; // authorized executor
bytes32 scope; // permitted action set
uint64 issuedAt;
uint64 expiresAt;
bytes principalProof; // ZK proof of humanity
}Look at what is absent. No revocation nonce. No registry pointer. No monotonic epoch tied to the principal.
Without them, revocation is a promise, not a mechanism. A hijacked agent holds a cryptographically valid proof until expiresAt, and the principal has no unilateral way to void it. Every serious delegation design I have reviewed in two years carries an epoch counter for exactly this reason. Long expiry plus no revocation path equals a credential that outlives the trust that created it.
Trust is a variable you cannot hardcode.
The release is silent on the question that decides whether this product can legally exist in Europe.
Two architectures are possible. In the first, the bank receives a zero-knowledge proof: a boolean attesting humanity plus required attributes. Nothing else crosses the boundary. GDPR-compatible in principle, still requiring a DPIA and a lawful basis for capture.
In the second, the bank receives an iris hash or template derivative. That is Article 9 special-category data, and the exact fact pattern behind BIPA class actions in Illinois with statutory damages per violation. No European DPO signs that architecture.
World has published nothing about which one it built. Absent disclosure, the claim is unfalsifiable. Unfalsifiable claims are the cheapest thing in this industry.
Then the structural problem, which predates this announcement and survives it.
B2B contracts are signed by legal entities. Tools for Humanity signs, invoices, and fiat lands on the equity side. WLD holders receive ecosystem proximity.
Four information points in the release. Not one mentions the token. No pricing in WLD, no fee flow, no buyback, no burn, no on-chain settlement. If the verification API is a centralized endpoint banks call without touching a chain โ the path of least resistance in procurement โ this is a narrative catalyst with zero cash-flow linkage.
Two instruments point at the same story. Only one has a claim on the revenue.
Data does not lie, but it does not care.
The competitive framing is wrong by an order of magnitude. World is not competing with Humanity Protocol's palm-vein scan or Polygon ID's verifiable credentials. Those are rounding errors in bank procurement.
The incumbents are Jumio, Onfido, Persona, LexisNexis. They hold existing master service agreements, SOC 2 Type II reports, penetration test histories, reference customers. Banks buy audit trail and SLA, not decentralization. The platform threat is Apple and Google shipping identity wallets at the OS layer with zero marginal hardware cost.
Against that field, World brings a hardware dependency, a contested biometric modality, and regulatory history across multiple jurisdictions. That is not a competitive position. That is a procurement objection list.
They built a palace on a fault line.
Now the part the bears will miss.
This is the first demand source World has ever had that is not purchased with inflation. Every prior verification was economically reflexive: price up, incentive attractive, volume up; price down, volume collapses. A bank paying fiat for an API call breaks the loop. Small, slow, unglamorous โ and the only structurally new thing in four years.
Second, the design partners are asset managers, not retail banks. Probably not a mistake. VanEck and Pantera sit at the RWA and compliant-DeFi end, where KYC gating is a hard requirement and the buyer is a fund, not a branch network. A smaller wedge. A more plausible one. Fund-level KYC runs a nine-month cycle, not twenty-four.
Third, regulatory legitimization is real. A regulated counterparty's supplier-risk review is an audit World cannot buy. Passing it converts the company's largest liability into a documented remediation narrative. That carries option value in every jurisdiction where it is currently unwelcome.
None of this makes the announcement true. It makes the direction non-trivial.
Demand four things before believing any of it.
FAR and FRR โ false accept and false reject rates on the Orb, third-party evaluated. Identity without error rates is a demo.
The ZK boundary. Proof or template. One word answers the European question.
Integration depth. Optional signal in onboarding, or mandatory step. Optional is worth nothing. Mandatory is worth a moat.
Revenue linkage. Does any dollar reach WLD.
Until those exist, this is a press release with a logo row and an unsourced number. The question is not whether World gets into a bank. The question is what the bank does the first time the identity vendor's regulator calls โ and whether the credential it depends on can be revoked before the answer arrives.