Imagine a procurement agent that spots a shortage of a critical component, identifies an alternative supplier, negotiates terms, places the order and initiates payment. No human touches the transaction. The humans set the rules; the agent executes.
AI agents are moving beyond generating content and assisting humans. They are beginning to act: negotiating with suppliers, purchasing goods and services, entering into agreements, initiating payments, arranging logistics and, increasingly, transacting with other AI agents. Analysts call this agentic commerce, and adoption is moving quickly: Gartner predicts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. The same pattern is emerging across procurement, financial services, supply chains, insurance and trade.
The constraint is no longer capability ,the capability is available and proven. The constraint is that the party on the other side of the transaction has no reliable way to establish who an agent is or what it may do.
The identity problem
Go back to the procurement scenario, this time from the supplier’s side. Its sales system receives an order for 40,000 units from something calling itself “the purchasing agent of Company X”. Before it commits inventory and extends credit terms, it needs to establish who the agent is, which organisation it represents, whether that organisation has authorised it to act and within what limits, and whether any of this can be verified independently in the seconds available.
A human buyer would settle those points through context. A business card, an email domain, a phone call to head office, a trading history, maybe a person-to-person relationship. A human also carries personal accountability. Misrepresent your employer or commit fraud and there are legal and employment consequences, which is part of what keeps human commerce honest.
An agent has neither the context nor the accountability. It cannot be fired or prosecuted, and fears neither. So verification has to do the work that context and consequences do between people, and accountability has to flow through identity to the organisation behind the agent.
The organisation is the principal
An AI agent is not an independent economic actor. It acts on behalf of an organisation. The agent may be the actor, but the organisation is the principal.
The supplier in the earlier example shows why the distinction has commercial weight. It does not care which model the agent runs on or which platform hosts it. It cares which legal entity stands behind the order, whether that entity authorised the agent to place it, and who pays if something goes wrong. Any workable trust model for autonomous commerce has to start with organisational identity, then attach agents and their permissions to it.
What verification looks like in practice
An autonomous transaction needs an unbroken chain: identity, representation, authority, action.
Take a second example, this time in payments. A corporate treasury agent instructs the company’s bank to pay a supplier $250,000. Today, banks catch fraudulent payment instructions partly through human checks: callbacks, dual approval, a familiar voice. When the instruction comes from software, the bank instead needs to verify a chain of claims. This instruction came from Agent A. Agent A represents Company X, identified by its LEI. Company X has authorised Agent A to initiate payments up to a defined limit, to approved beneficiaries. Each link is cryptographically verifiable, so the bank’s systems can accept or reject the instruction in milliseconds.
An API key cannot carry that chain. A key proves possession of a secret, and says nothing about which legal entity is behind the agent or what that entity has authorised.
Where the LEI and vLEI fit
If agents must prove which organisations they represent, we need a globally recognised way of identifying those organisations. That part, at least, already exists.
The Legal Entity Identifier is an ISO-standard identifier for legal entities, referenced in regulations worldwide and anchored in the Global LEI Index. The verifiable LEI (vLEI) extends it into digital interactions: a cryptographically verifiable credential that lets an organisation prove its identity and issue credentials to its authorised representatives, human or machine, under a governance framework GLEIF already operates. GLEIF has demonstrated vLEI credentials assigned to AI agents, allowing a counterparty to verify an agent’s organisational origin in real time.
What identity alone does not solve
A caveat before going further: identity solves part of this problem, not all of it.
Verifiable identity tells the counterparty who the principal is and what mandate the agent claims. It does not settle questions of liability. If an agent exceeds its mandate, contract law in most jurisdictions has not yet determined when the principal is bound. Credentials create the evidence trail; they do not decide the case.
Nor does identity eliminate fraud. Credentials make impersonation much harder, but a legitimately credentialed agent can still be manipulated into bad decisions, and a compromised credential is dangerous until revoked. Governance questions also remain inside the organisation: who grants an agent its mandate, who reviews it, how quickly can it be withdrawn.
None of these problems is solved by a credential alone. Each becomes workable, though, once the parties can establish who is acting for whom, which is why identity comes first.
The network problem
Holding a credential achieves little if the receiving system cannot recognise it. Suppose a logistics agent books freight capacity and then tries to bind cargo insurance with an insurer’s underwriting agent. If the insurer’s systems have no way to verify the logistics agent’s credential, the transaction falls back to email and human review, and the efficiency gain disappears.
Autonomous commerce therefore depends on adoption across the systems transactions already run on: payment networks, banks, ERP and procurement platforms, technology providers. HTTPS is the useful precedent. Encryption became ubiquitous because browsers, hosting providers and certificate authorities made it the default everyone expects, not because anyone was compelled. The aim is an ecosystem where verified identity makes autonomous transactions easier, safer and cheaper than the unverified alternative.
Regulators as enablers
The regulatory question is usually framed as how to stop AI agents from doing dangerous things. A more productive question is what infrastructure would let them safely do useful things.
Regulators and standards bodies can help establish common identity standards, interoperable credentials, and clear rules on authority and representation, including the liability questions above. Handled well, regulation gives organisations the confidence to transact with agents at all. That produces more trusted autonomy, not less.
When machines transact with machines
Over the next few years, agents will discover, identify, verify, negotiate, agree, transact and settle with each other, while humans set the rules rather than approving each action. For that to work, trust itself has to be something software can check.
Once machines act on behalf of organisations, identity becomes infrastructure. Organisations that establish verifiable identities for their agents will be in a position to take part, and the ecosystems that make those identities interoperable will decide how far this scales.
Trust does not stand in the way of autonomy. Trust enables it.
The LEI is the global standard for organisational identity, and the vLEI extends it into verifiable digital interactions. To learn how your organisation can prepare, contact RapidLEI.
References
- Gartner, Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, August 2025.
- GLEIF, Why AI Agents Need Verifiable Organizational Identity, GLEIF Blog.
- GLEIF, Introducing the verifiable LEI (vLEI).
- GLEIF, vLEI Ecosystem Governance Framework.
- GLEIF, ISO 17442: The LEI Code Structure.
- GLEIF, Global vLEI Hackathon: How Affinidi and GLEIF Are Accelerating Privacy-Preserving Identity for the AI Era, GLEIF Blog.
- Biometric Update, GLEIF, BIS test vLEI as trust layer for cross-border open finance, June 2026.
