AI Agents Are Learning to Pay. The Compliance Stack Is Still Built for Humans.

Published: July 27, 2026
Category: Technology / Agentic Commerce
Reading Time: 6 minutes

---

Natural, a startup building payment infrastructure for AI agents, announced a $30 million Series A this week, led by Forerunner. The round brings its total funding to $40 million and puts it in direct competition with Stripe, which updated its Link wallet for autonomous AI agents in April, and with Skyfire Systems, which is building agent-payment rails using stablecoins.

This is not a niche fintech story.

It is the first clear signal that the next major compliance architecture problem is no longer theoretical. AI agents can now identify vendors, compare prices, negotiate delivery, and — increasingly — initiate payments and execute trades. The financial rails underneath them were designed for humans clicking "confirm."

That gap is the story.

---

What is actually changing

Traditional payment systems — credit cards, ACH, wire transfers — assume a human originator. The identity layer, the authorisation layer, and the monitoring layer are all built around that assumption.

When Robinhood announced in May that AI agents can now trade stocks on its platform, and when Stripe opened Link to autonomous agents in April, the operational question stopped being "will AI agents transact?" and became "who is responsible when they do?"

Natural's pitch is straightforward: build an orchestration layer that lets AI agents move and store funds, pay vendors, collect payments, and transact with other agents. Its CEO, Kahlil Lalji, told TechCrunch that "agentic payments are going to be structurally the most important problem in the space" and that payment volumes could be "two or three or four orders of magnitude greater" than today if transactions happen at computer speed rather than human speed.

That scale is precisely why the compliance question is urgent.

---

Three compliance gaps that have not been closed

1. The identity gap

KYC and CDD are designed around natural persons and corporate entities. An AI agent is neither. It may act on behalf of a person, a company, a pool of users, or another AI agent.

Current systems have no standard data model for "agent acting under delegated authority." The result is that either the agent transacts under the human customer's credentials — flattening the audit trail — or the transaction is blocked because the system cannot resolve who the obliged entity is.

2. The authorisation gap

In most jurisdictions, an AI cannot be a legal party to a contract or a payment. The authorisation must still come from a human or a corporate entity. But when an agent is making hundreds of low-value decisions per hour based on learned rules, the traditional model of "human approves each transaction" collapses.

The legal concept of delegated authority exists, but it has never been stress-tested at agentic scale. What level of pre-authorisation is sufficient? Who bears liability for an agent-initiated payment that turns out to be fraudulent? These questions do not have settled answers.

3. The surveillance gap

Transaction monitoring systems are trained on human behaviour: time-of-day patterns, velocity limits, geographic concentration, device fingerprints. AI agents behave differently. They may transact continuously, across jurisdictions, at machine speed, and through APIs rather than browsers.

A monitoring stack that flags "unusual human behaviour" will either miss agent-based risk or generate so many false positives that it becomes unusable. The underlying models need to be rebuilt around autonomous-agent patterns, not retrofitted.

---

Why this matters now

The Natural funding is a market signal. Investors are betting that agentic payments will become a distinct infrastructure layer, separate from consumer and B2B payments. That bet only makes sense if the volume materialises. And the volume will only materialise if regulators, banks, and compliance technology providers accept that the current stack is inadequate.

This is not a call for more rules. Existing AML, consumer protection, and payment-services frameworks already apply. The problem is that the technical implementation of those rules assumes a human at the centre.

Rebuilding that implementation — identity proofing for non-human actors, delegated-authorisation records that survive audit, surveillance models for machine-speed transactions — is the compliance technology challenge of the next five years.

---

The observation that matters

We have seen this pattern before. Mobile payments, e-commerce marketplaces, stablecoins, and instant settlement each outran the compliance infrastructure built for the previous era. In every case, the institutions that survived were the ones that redesigned their compliance stack for the new actor, not the ones that treated the new actor as an edge case.

AI agents are the next actor. The compliance stack needs to stop assuming a human is clicking "confirm."

---

UWAY provides compliance infrastructure for FinTech, Web3, and digital asset businesses.

#AIAgents #FinTech #RegTech #Compliance #Payments #AML #DigitalAssets #UWAY