World Shipped a Human Check on October 18. Banks Still Need Four More Answers

I've spent the better part of a decade reading bank onboarding documentation, first on the equities desk, then on the credit side, and now here. There's one line that shows up in nearly every institution's identity program: authentication isn't authorization. World launched its banking and finance offering on October 18 that answers the first question rather well. The second question it doesn't answer at all, and that's where the money actually moves.

That's not a knock on the launch. It's the whole point of getting the plumbing labeled correctly before anyone builds a payment rail on top of it.

Three proofs, three different jobs

World's offering leans on World ID, and World ID isn't one thing. It's three credentials wearing the same badge. Proof of Human runs through Orb verification and establishes uniqueness, meaning this is one person and not the same person counted twice. Proof of Presence is a selfie-based check, which tells you someone showed up at the moment of the request. Proof of Attributes confirms the holder possesses a supported government document.

Read those again and notice they're answering completely different questions. Uniqueness doesn't tell you who. Presence doesn't tell you who. Attributes tell you who, but only as reliably as the document does, and forged documents have been a growth industry since the invention of paper.

Here's the gap that matters for a bank. A unique human can be an unauthorized human. Your branch teller is a verified, unique, present human being, and she still can't approve a $4 million wire at 11pm on a Sunday. She has identity. She doesn't have permission. Those are separate systems, and they should stay separate, because collapsing them is how you get a credential that means everything and therefore nothing.

The comparable in TradFi is the signing authority matrix inside a corporate treasury. Identity gets you through the door. The matrix decides which door you can open and how wide. No compliance officer on earth would merge those two documents into one, and yet the crypto industry keeps trying to do exactly that with a single credential and a nice landing page.

There's a technical reason to be careful, too. NIST's digital identity guidance, the 800-63-4 revision, treats a biometric characteristic as insufficient on its own for the authentication combinations it specifies. That's not a verdict on World's implementation. It's a reminder that standards bodies looked at this problem years ago and concluded that one factor, no matter how elegant the hardware, isn't a control system. It's a component of one.

In traditional markets, this would be called separation of duties, and it exists because humans have been forging credentials since credentials were invented.

Agents break the model faster than banks can fix it

Now layer AI agents on top and the gap widens into something structural.

Picture an assistant paying a supplier invoice. A proof that a real customer stands behind the request is genuinely useful. But the service still needs to know which account the agent can draw on, what the cap is, who it's allowed to pay, and when its authority expires. It needs a revocation path that doesn't depend on the agent cooperating with its own shutdown. And it needs a record of past consent that doesn't quietly convert into a permanent blank check.

Who owns that control? The bank holding the account, the developer who built the agent, or the identity provider that issued the credential? Nobody has cleanly answered it. The launch materials describe the credential and stay quiet on the liability chain, which is understandable for a product announcement and unacceptable six months from now.

Can you revoke a delegated payment permission as easily as you can revoke a session token? Today, probably not. That's the part of agentic finance nobody wants to demo on stage.

Transaction risk screening stays a separate layer as well. Knowing the participant and evaluating the destination are complementary tasks, not substitutes. Fraud screening ahead of stablecoin settlement is already becoming table stakes across the industry, and it exists precisely because identity doesn't travel with the money. The address doesn't know who sent it.

Privacy needs the same precision. A proof can reduce how much data gets shared at one verification step while the surrounding service still logs transaction history, account details, and device fingerprints. Customers should look at the entire data flow rather than the marketing slice of it. Strip away the jargon and it's a data governance question, and regulators are going to ask it in writing.

What this means, and what to do about it

The market reaction will be noisy and mostly wrong. World's token is going to trade on narrative, because it always does, and narrative is a poor pricing mechanism for infrastructure. Crypto is pricing in what equities haven't, namely a future where human verification is a paid utility layer with real recurring revenue. An equity analyst would ask a blunter question than any crypto trader: what's revenue per verified user, and what's the retention rate twelve months out?

My view, and it's a reasonably firm one: the winner here isn't whoever verifies the most humans. It's whoever owns the authorization layer sitting on top. Verification is commoditizing fast. Uniqueness proofs will get cheap, and governments may eventually provide them for free through digital ID programs already on the drawing board. Authorization, the messy business of limits, delegations, revocation, and audit trails, is where the sticky, high-margin, defensible work lives.

For anyone trying to model this, the Sharpe ratio tells a sobering story. Identity tokens have historically delivered venture-style upside with none of the contractual protections, so a treasury that can earn a decent yield in short-dated paper is taking a genuinely bad risk-adjusted trade by buying the narrative instead of the cash flow. The comparable in TradFi is a pre-revenue company trading at a growth multiple. Sometimes that works. Usually it doesn't.

Second opinion, slightly hotter: banks should run World and similar credentials through the same vendor risk framework they use for any identity provider, not treat it as a crypto experiment with a special budget line. Ask which specific fact each credential proves. Ask whether that fact feeds the authentication decision or the authorization decision. Ask what happens when the Orb goes offline, when a document format changes, or when a customer disputes a transfer the agent made at 3am. If the vendor can't separate those answers cleanly, you're not buying verification. You're buying a logo.

None of this makes the October 18 launch a mistake. It makes it an opening bid. World has proved something narrow and genuinely useful, and narrow proofs are how every decent identity system starts. The financial system needs something broad and operational, and the clock on closing that distance is running at the speed of AI agents that don't sleep, don't ask permission twice, and won't wait until Monday morning for a human to sign off.