Artificial Intelligence (AI) and Organizational Identity

As AI agents begin to find services, exchange information and initiate payments on behalf of organizations, a practical question arises: How can another party verify which organization stands behind an agent and what that agent is authorized to do? The Legal Entity Identifier (LEI) provides a globally standardized identifier linked to verified organizational reference data. The verifiable LEI (vLEI) extends this foundation into digital interactions, enabling counterparties to computationally verify organizational identity and the authority of people acting on its behalf.

GLEIF is examining how this infrastructure could support future approaches to delegated authority for AI agents, particularly in payments. Alongside this emerging area, standardized LEI reference data already supports AI-based classification of legal entities. These applications address different needs: verifying authority when AI agents take action, and improving the organizational data used in AI-driven analysis.

Agentic AI in Payments: Verifying Identity and Delegated Authority

An organization may delegate payment tasks to software, but a counterparty still needs evidence of who authorized the agent, what actions it may take, and how those permissions are checked. The LEI identifies the legal entity involved. Existing vLEI credentials enable computational verification of organizational identity and the roles and authority of people acting for that entity. Connecting a specific AI agent to an organization and its permitted actions requires additional delegation evidence and controls at the time of the transaction.

GLEIF’s September 2026 working paper, “Agentic AI in Payments: Establishing Interoperable Trust and Control”, explores one possible approach. The paper examines a proposed Agent Mandate Credential (AMC) that could connect an AI agent’s cryptographic identifier to a verified organization and an authorized person, while specifying permitted tasks, transaction limits and validity. For example, a corporate treasury officer could authorize an agent to execute defined supplier payments; the payment service provider could verify the authorization chain and apply transaction-time controls before executing. The paper also examines revocation, audit records, privacy and interoperability. The AMC remains a proposal for further research and testing. It is not currently a credential issued within the vLEI ecosystem or a finalized industry standard.

AI-Powered Data Analysis: Legal Entity Name Understanding (LENU)

Legal forms vary across jurisdictions, languages and naming conventions, making them difficult to classify consistently. The Entity Legal Forms (ELF) Code List, based on ISO 20275, assigns a four-character alphanumeric code to each entity legal form. GLEIF collaborated with Sociovestix Labs to develop Legal Entity Name Understanding (LENU), an open-source machine learning tool that predicts an entity’s legal form from its name and suggests the corresponding ELF code. LENU uses standardized LEI reference data to train jurisdiction-specific models and can support:

• More consistent legal-form classification across languages and naming conventions.
• Better entity matching and data integration when records use different labels.
• Less manual effort in preparing legal-form data for downstream processes.
• Data quality reviews when suggested ELF codes differ from those in existing LEI records.

These machine-readable classifications help organizations prepare more consistent entity data for analytics, onboarding and other digital processes. The following GLEIF Explains video introduces LENU.

LENU is an open-source Python library available on GitHub, with jurisdiction-specific models also available on Hugging Face. Its transformer-based models can process entity names across languages and jurisdictions.

Where GLEIF identifies clear discrepancies between LENU predictions and existing LEI records, it can send a data challenge to the relevant LEI Issuer for verification and correction where needed. Updated reference data can then inform subsequent model versions.

GLEIF, the University of St. Gallen and Sociovestix Labs published the scientific research paper “Transformer-based Entity Legal Form Classification”, which examines the approach and its potential to support data standardization and entity matching.

For an overview, read our latest blog ‘When AI Agents Transact, Who Stands Behind Them?’. For the wider payments context, see Cross-Border Payments