When AI agents transact, who stands behind them? Why verifiable organizational identity is critical to agentic payments

Businesses everywhere are rapidly increasing their dependence on AI agents. Happily, the legal principles that assign responsibility remain robust. What’s missing is the infrastructure to prove, at machine speed, which organization stands behind an agent and who authorized it to act. Drawing on a new GLEIF discussion paper on agentic AI in payments, this explores how a trust layer of verifiable organizational identity addresses the issue.

---

---

Today's AI agents are doing far more than drafting text and staffing chat boxes. They are discovering services, negotiating terms, and initiating payments, increasingly with other agents. Major payment networks and infrastructure providers are moving to support agent-initiated transactions, highlighting that this is no longer a theoretical discussion; it's an operational one.

The scale of the shift is unprecedented. Gartner projects that up to 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% a year earlier. More than 60% expect to deploy agents within two years, it says, the steepest adoption curve of any emerging technology it tracks.

In financial services specifically, the Cambridge Centre for Alternative Finance reports that, by early 2026, 52% of institutions were at least piloting agentic AI, with many further along.

The whole addressable market is, of course, much wider, reaching into payments, trade and supply chains, digital assets, and government services. Yet, across all these sectors, the same questions persist: Who should give an organization's agent authority to act, and how can a counterparty verify that authority has been granted?

For those of us who work on organizational identity, the pattern is familiar. A new class of actor emerges and only then are the systems built to identify and hold that actor accountable. In the agentic world, the problem is not the absence of law. It is missing proof.

When an agent makes a payment that no one expressly approved, the question of who is responsible for it still feels new, yet in most cases it really isn't. The legal frameworks that allocate liability are, for the most part, already mature. What they can't do, however, is provide a practical way to demonstrate delegated authority, establish legal standing and, when a dispute arises, produce reliable evidence.

In this way, the bulk of the work that now needs to happen is in the developer's domain, not the lawmaker's. The challenge is to translate established legal principles into technical and operational practices that agents can execute. More specifically, each agent needs to carry machine-readable proof of three distinct things: the legal identity of the organization it acts for, the authority it has been delegated, and the boundaries of that authority. When those can be verified computationally, accountability stops depending on trust and starts depending on evidence.

Much more than Know Your Agent

Much of the current effort to identify agents stops at the device or the individual credential. That part still matters, of course, and is developing quickly. But it does not answer the question a bank, a supplier, or a regulator actually needs answered: on whose authority is this agent acting, and is that authority real?

This is where the verifiable LEI (vLEI) fits. The Legal Entity Identifier (LEI) already answers the 'who is who?' and 'who owns whom?' questions for more than 3.4 million legal entities worldwide. The vLEI, its digital counterpart, extends that verified organizational identity into the digital domain, enabling counterparties to computationally verify the identity, authority, and role of a person or agent acting on behalf of a legal entity. It sits above device and agent identity as the organizational trust layer, complementing work at the device and individual credential levels.

A hardware key can prove that a specific agent is running on a specific device. It cannot, on its own, prove that the agent represents a licensed importer in Rotterdam, or that the person who authorized it holds a role that carries the requisite authority. The vLEI makes this possible by using a cryptographic trust chain that traces back to a verified individual within a verified organization. Put another way, accountability is hardwired into each agentic transaction.

Making digital organizational identity work in practice

Agentic systems don't behave like working people. Typically, agents perform narrowly defined tasks, in accordance with specific contractual arrangements, over short periods. This is because the people and organizations responsible for them need to issue, monitor, and revoke authority frequently, and sometimes almost straight away.

Governing agents with a 'partitioned authority' model – controlling which agent can do what, under whose mandate, for how long, and according to what limitations - is much more efficient than arming each agent with its own distinct identity. A credential model built on digitally verifiable organizational identity can be designed precisely with these factors in mind, enabling the agent's authority to accompany the credential and be verified when the transaction is initiated, rather than reconstructed afterward.

Cross-border payments are the proving ground

While the potential use-cases for this organizational identity model are virtually limitless, the world of cross-border payments will likely be the first beneficiary. Here, multiple jurisdictions, providers, and rulebooks routinely collide in a single transaction.

For agent-initiated cross-border payments, an agent's organizational credential needs to be machine-readable by counterparties and payment providers across jurisdictions, without depending on bilateral trust arrangements or platform-specific registries. The systems must be able to check whether the transaction corridor is permitted, whether the counterparty is allowed, whether the mandate is still valid, and whether the organization standing behind the payment can be independently verified.

Happily, the global financial system's migration to ISO 20022 structured data continues at pace, meaning that this level of identity precision can be incorporated into the payment message itself, and could soon become a basic expectation more or less everywhere.

A public good, extended

None of this replaces what the Global LEI System already does. The system was established as a broad public good with an initial focus on capital markets. It is evolving into a global Digital Public Infrastructure for business identity, anchoring organizational identity across payments, supply chains, digital assets, and government services. Supporting accountable agentic activity is a natural continuation of that role.

The agentic economy will reward the ecosystems that hardwire trust into their operations. Open, standardized, verifiable organizational identity is one of the pieces that makes that trust possible. GLEIF will continue working with standards bodies, regulators, and industry to help implement it. GLEIF's full discussion paper on agentic AI in payments sets out the mechanisms in more detail, and invites further discussion and feedback.

---