Financial Crime Post Money Laundering

I’ve got a crime to solve, and I need your help cracking it. Here’s the case:

It’s Tuesday. The document is genuine. The face is live. The screening comes back clean, and the customer was verified months ago. The payment instruction arrives correctly signed, the amount sits comfortably within limits, and the beneficiary is already approved. And the audit trail? Honestly, it’s beautiful.

Six weeks later, however, the bank discovers that no human ever approved that transfer. Not the customer. Not an employee. No one. 

So how did this happen? Let’s look at the evidence: Every signature in this transaction was real and every check passed. So, where’s the crime? We’ve all heard the cliche, “follow the money.”  But in this case, it isn’t the money. The money moved where it was directed. The crime is in the instruction. Somewhere between a verified human and a cleared payment, authority changed hands without anyone sanctioning it, and every system along the way waved it through as legitimate.

Following the money has been the going motif in financial crime fighting for the past thirty years. The entire compliance stack, from transaction monitoring to source of funds checks, is built to trace value and spot the moment dirty money slips into a clean suit. But our Tuesday transfer is hiding where instinct never looks.

I call it authority laundering: making unauthorized power look like it was legitimately handed over. And the discipline built to catch it, Know Your Agent, or KYA, is about to become as fundamental to financial institutions as KYC is today.

Who’s clicking?

AI agents are now opening accounts, negotiating purchases, and sending payments on behalf of people and companies. The payments industry isn’t waiting around. Agentic payment protocols backed by the major card networks already let software carry cryptographically signed instructions, called mandates, that prove a user authorized a transaction.

Banks are catching on to what’s happening. And the threat isn’t hypothetical anymore. In 2025, for the first time in its 25-year history, the FBI’s Internet Crime Complaint Center gave artificial intelligence its own section in the annual report. It logged more than 22 thousand AI related complaints and about $893 million in losses, driven by voice clones, fake identification documents, and deepfake video. Some of those attacks weren’t even after money. The FBI documented deepfake job interviews where the real goal was getting hired, getting credentials, and getting inside.

Authority Laundering

Money laundering cleans up dirty money by associating it with a believable business, a paper trail, books that balance. Authority laundering does the same thing for power. If you know the classic AML playbook, you’ll recognize every move. It’s just been rebuilt around delegation instead of cash.

1. Layered delegation. A human authorizes agent A. Agent A delegates to agent B. Agent B spins up agents C through H across three platforms and two jurisdictions. Look at any single link and it checks out. Look at the whole chain and you can’t tell who’s really in control, which is exactly how layered shell companies hide beneficial ownership. The difference? Setting up a shell company takes time. And lawyers. Spinning up a shell agent takes one API call and a few milliseconds.

2. Mandate mules. Money mules rent out their bank accounts. Mandate mules will rent out their signature: real, verified people paid to approve mandates for agents they don’t control. The recruitment ads will read just like today’s money mule ads, because it’s the same crime with a new asset. Put one verified human signature at the top of a delegation chain and everything underneath it comes out looking clean.

3. Synthetic principals. The deepfake onboarding attacks we’re already seeing become the way in. A synthetic identity passes KYC, becomes a “verified human,” and then hands authority, perfectly legitimately, to a whole fleet of agents. The fraud doesn’t happen at the transaction. It already happened at onboarding, and every agent action afterward inherits that false legitimacy.

4. Zombie mandates. Authority that should be dead but isn’t. A mandate revoked by a principal that remains “live” in downstream services with cached credentials. An agent whose ownership quietly changed hands after it was verified. Yesterday’s legitimate delegation turns into today’s unauthorized power, and the paperwork still smiles back at the auditor.

Any one of these four could be behind our fraudulent Tuesday transfer. And that’s what makes authority laundering so dangerous. It doesn’t beat your controls. It recruits them.

The KYC Failure

KYC answers one question, at one moment: is this person who they say they are at onboarding? In the Tuesday transfer, KYC did its job perfectly. But it was answering a question no one’s asking anymore. KYC was never built for customers who have no birthday and no face, can be copied a million times before lunch, move at machine speed, and get all their legitimacy from somebody else.

Verifying an agent the way we verify a person is like checking a car’s registration and never asking who’s behind the wheel. The agent’s own identity is the least interesting thing about it. What matters is the authority it’s carrying and where that authority came from.

Introducing: Know Your Agent

Done right, KYA isn’t about authenticating the agent. It’s about verifying a chain of accountability, and it comes down to four questions.

1. Who’s the principal? It’s the verified human or legal entity at the start of the chain, and it’s where identity verification stays irreplaceable: document authentication, biometrics with deepfake and injection detection, and more and more, wallet based credentials. Think of it as the reality check, proof that a real human being is standing at the beginning. Every mandate in the world is worthless if the identity that signed the first one was synthetic.

2. What’s the mandate? The specific, verifiable instruction the principal actually gave: how much can be spent, on what kinds of actions, and for how long. Cryptographically signed, tamper proof, and fully auditable.

3. What’s the scope? The boundaries, checked continuously instead of assumed. If an agent authorized to book travel starts sending money to new beneficiaries, that’s not a technicality. That’s the agentic version of a transaction monitoring alert.

4. How does revocation work? It’s the least glamorous question, and it’s the one that will decide real cases. When the principal pulls authority, how fast does everyone downstream find out? It’s offboarding for delegated power, with proof.

Principal, mandate, scope, revocation. Any institution that can answer all four for every agent touching its systems gets to say the one sentence that really matters: every agent action in our environment can be traced, in one hop, to a human who answers for it.

Demand before regulation

Here’s my prediction. AMLR arrives in July 2027, and we’ve already been working through what it means. But in that regulation, “agent” means something else entirely, and formal KYA rules are realistically years away. Still, obliged entities won’t wait, because they face liability regardless of regulation. When an agent transaction goes wrong, the institution that allowed it is the one held accountable, facing financial and reputational damages.

So the KYA demand will become a new norm through contracts, network rules, and insurance underwriting, not legislation. Banks will ask platforms for verifiable mandates the same way they ask customers for ID. By the time a regulator actually writes “Know Your Agent” into guidance, it’ll already be table stakes.

For compliance and product leaders, that flips the planning question. It’s no longer “when’s the deadline?” It’s “when do our counterparties start asking?” And the honest answer is that some already are.

Take action

5 actions you can take today, without waiting for anyone to finalize a standard.

1. Check your traffic’s heartbeat. Know which of your traffic is human and which isn’t.

2. Nail the ground floor. Every delegation chain starts with a human passing identity verification. That makes synthetic identities and deepfake onboarding the root vulnerability of the entire agent economy. If the principal can be faked, nothing downstream can be trusted.

3. Know who authorized every action. The mandate, its scope, its chain, and whether it’s been revoked, all available on demand.

4. Write your revocation procedure now. Not after the first incident.

5. Follow the authority.

Talk to us about agent era readiness. AU10TIX verifies the humans at the origin of every delegation chain, with deepfake resistant biometrics, document forensics, and audit ready evidence trails built for what comes next.