Know Your Business Agent: Why AI Agents Will Need Company Verification Before They Can Do Business

AI Agents Are Becoming Economic Actors

The first generation of enterprise generative AI primarily assisted employees. It drafted emails, summarised documents and answered questions. The software could influence a decision, but a human usually executed the action.

Agentic systems are changing that model. An AI agent can receive an objective, select tools, retrieve external information and complete a multi-step workflow with limited intervention.

In enterprise environments, agents are being designed to:

• Identify and compare suppliers.
• Enrich company and customer records.
• Review invoices and payment requests.
• Perform business-verification checks.
• Assess financial and ownership information.
• Request quotations or commercial terms.
• Interact with other agents and external applications.

These are not merely content-generation tasks. They are operational activities that may affect contracts, payments, access permissions, procurement decisions and regulatory obligations.

The more authority an agent receives, the more important this binding becomes. A research assistant presents relatively limited risk. An agent permitted to alter supplier details or initiate a large transaction presents a very different risk profile.

Why Agent Identity Alone Is Not Enough

Identity infrastructure can answer important questions:

• Which agent is making the request?
• Which person authorised it?
• Is the agent using valid credentials?
• What permissions were delegated to it?

Enterprise transactions require additional business context:

• Which legal entity operates the agent?
• Is that company active and registered?
• Who ultimately owns and controls it?
• Is the agent acting for the parent company or one subsidiary?
• Is the business financially and operationally eligible?
• Has anything material changed since the agent was approved?

Neither layer should make the complete decision independently. Agent identity, company evidence, delegated authority, transaction intent and enterprise policy must be evaluated together.

Agent identity establishes which software is acting. Company intelligence establishes which organisation is accountable.

What Is Know Your Business Agent?

KYBA extends conventional business verification into autonomous workflows. It does not treat the agent as an isolated piece of software. It treats the agent as a digital representative of an organisation.

A mature KYBA response could include:

KYBA is not yet an established international standard. It is a category that enterprises, identity providers and company-intelligence platforms will need to define as autonomous systems enter higher-risk commercial workflows.

The Minimum Viable KYBA Standard

A category becomes useful only when buyers can distinguish a genuine implementation from a renamed company lookup. At minimum, a KYBA service should satisfy six requirements.

• Persistent agent identity.
  The service must identify a durable agent or workload, not merely an anonymous browser session.
• Exact legal-entity binding.
  The agent must be matched to a legal entity through official registration, tax or other authoritative identifiers.
• Independent relationship verification.
  The organisation-to-agent connection cannot rely solely on a company name entered during registration.
• Recorded delegated authority.
  The system must retain the permitted actions, value limits, approving principal and expiry of authority.
• Source-linked evidence.
  Material company facts must include the registry, filing or authoritative source and the time retrieved.
• Event-driven reassessment.
  Material changes to the company, authority or agent must trigger review, restriction or revocation.

This definition prevents a basic company-name search or a permanent trust badge from being presented as a complete KYBA control.

How KYBA Could Work

Consider an AI procurement agent attempting to onboard a supplier. Before the agent can recommend approval, a KYBA workflow could perform five stages of verification.

The company-intelligence provider should not impose one universal definition of trust. It should establish the corporate facts. The receiving enterprise should retain control over its own risk policy.

A Worked KYBA Transaction

Consider a procurement agent requesting enterprise pricing and authority to negotiate a three-year data licence.

{
  "agent_verified": true,
  "legal_entity_verified": true,
  "ultimate_parent": "ABC Global Holdings plc",
  "existing_group_agreement": true,
  "delegated_limit_gbp": 250000,
  "requested_commitment_gbp": 310000,
  "decision": "human_approval_required",
  "reasons": [
    "Requested commitment exceeds delegated authority",
    "Existing group agreement requires account-owner review"
  ],
  "evidence": {
    "company_registry_source": "linked",
    "ownership_last_checked": "2026-07-28T07:14:00Z"
  }
}

Enterprise Use Cases for KYBA

Who Owns KYBA Inside the Enterprise?

KYBA crosses several established control functions. This is both its value and its adoption risk: everyone may agree the problem matters while no single team owns the budget.

Why Official Registry Data Matters

Large language models can reason about corporate information, but they are not the authoritative source of that information.

A model can explain a balance sheet or summarise an ownership structure placed in its context. It cannot independently guarantee that the latest directors, shareholders, financial statements or legal status are correct at the time a decision is made.

Corporate facts change continuously:

• Companies are incorporated, dissolved and restored.
• Directors and legal representatives change.
• Businesses are acquired or moved into new corporate groups.
• New accounts and financial statements are filed.
• Ownership and beneficial ownership records are updated.
• Companies enter insolvency or liquidation.

An autonomous decision should therefore retrieve current evidence rather than rely solely on model memory, web summaries or a static company profile.

The AI should determine what information the workflow needs. The company-intelligence layer should establish what the available official evidence says.

Provenance matters as much as the answer. Enterprise users need to know which source supplied the information, when it was retrieved and whether the conclusion can be reproduced later.

Agent Identity Is Becoming a Standards Problem

KYBA is not emerging in a vacuum. Public standards initiatives are beginning to address secure agent identity, interoperability and authorisation.

NIST is organising an agent-standards initiative

On 17 February 2026, the US National Institute of Standards and Technology launched its AI Agent Standards Initiative. NIST said the programme would foster industry-led standards and open protocols, with specific research into agent security and identity so agents can act securely on behalf of users and interoperate across the digital ecosystem.

OpenID is developing authorisation profiles for agent workflows

On 15 June 2026, the OpenID Foundation's AuthZEN Working Group approved official Working Group Drafts for the Access Request and Approval Profile and MCP Tool Authorization profile. The work addresses cases in which a policy cannot yet authorise an action because approval, consent, delegated authority, attestation or another prerequisite is still required.

IETF participants are documenting agent authentication and delegation

An active July 2026 Internet-Draft on AI Agent Authentication and Authorization describes how existing standards such as OAuth 2.0 and workload-identity architectures could be applied to agent interactions, delegated authority, audit trails and monitoring.

These initiatives focus primarily on identifying agents, authenticating workloads and determining whether an action is authorised. They do not independently resolve the adjacent corporate question: which verified legal entity is commercially responsible for the agent?

Emerging standards can carry agent identity and authority. KYBA supplies the verified corporate identity those controls need to become commercially accountable.

The Agent-Trust Market Is Already Taking Shape

Identity providers, payment networks and internet-infrastructure companies are already developing parts of the agent-trust stack.

Experian Agent Trust

Experian announced Agent Trust on 30 April 2026. Its Know Your Agent framework uses Human-to-Agent Binding to create a persistent connection between verified consumers, devices and the agents acting on their behalf.

Experian also described an Agent Trust Token that evaluates identity and transaction risk, supported by an Agent Registry with dynamic trust scoring based on behaviour and other risk signals.

On 24 July 2026, Experian announced that Fastly had joined the ecosystem. The collaboration is intended to allow organisations to evaluate agent identity, delegated authority, intent and payment credentials at the network edge before a request reaches the underlying application.

Visa Trusted Agent Protocol

Visa's Trusted Agent Protocol provides a cryptographic framework through which merchants and site-protection providers can distinguish approved commercial agents from malicious bots or unidentified automated traffic.

The protocol is designed to carry linked, verifiable and time-bound credentials over existing web and API infrastructure, helping merchants recognise the agent and the commercial intent attached to the interaction.

Mastercard Agent Pay

Mastercard has developed Agent Pay around a Know Your Agent trust framework. Mastercard describes an agent as being identified, registered, authenticated and required to demonstrate verified intent before transacting.

These initiatives establish that agents need recognisable identities, permission controls and traceable transaction intent. However, they also expose the adjacent corporate question: which legal entity is accountable for the agent?

The Complete Enterprise Agent-Trust Stack

Over time, these signals could be presented through a machine-readable Business Agent Passport. The credential could include the persistent agent identifier, verified legal entity, relevant parent company, authorised domains, delegated permissions and references to the evidence supporting the binding.

The passport should not provide universal approval. Each receiving organisation would validate the evidence and apply its own policy to the requested action.

What a Business Agent Passport Could Contain

A portable credential could allow an agent to present its identity, legal entity and authority in a consistent format. The receiving enterprise would still validate the underlying evidence and apply its own policy.

KYBA Establishes Attribution, Not Legal Capacity

Verifying an agent, its user and its company does not automatically prove that the agent can legally bind the organisation to every contract or transaction.

This boundary strengthens rather than weakens the proposition. A credible KYBA service should state precisely which facts it verifies and leave legal and policy decisions with the organisation responsible for them.

Where Global Database Fits

Global Database already provides the company-verification and corporate-evidence capabilities required within a KYBA architecture, including legal entity resolution, ownership, group structures, financials and source-linked registry evidence. It does not currently market a standalone agent-trust layer for issuing persistent agent identities, binding agents cryptographically to organisations, managing delegated permissions or enforcing those permissions.

Its immediate role in the emerging ecosystem is therefore to verify the legal entity behind an agent and supply the corporate evidence used by identity, authorisation and policy-enforcement systems.

Global Database collects company information directly from official registries across more than 200 countries. Its current data estate covers more than 600 million company profiles sourced from over 400 official registries, with source attribution and last-verified timestamps.

Relevant KYBA outputs can include:

• Exact legal entity and registration status.
• Company registration, VAT, tax, EIN and LEI identifiers.
• Directors, officers and authorised representatives.
• Shareholders and beneficial ownership.
• Parent companies, subsidiaries and group structures.
• Filed financial statements and financial indicators.
• Registry source, filing reference and retrieval timestamp.
• Monitoring for changes in corporate status and control.

These capabilities can be consumed through several delivery models, including Regis, the Regis API and the Global Database MCP server.

The strongest positioning is not that Global Database is a database agents can search. It is that Global Database can become the corporate evidence layer used to establish which legal entity stands behind a commercial agent.

Agent-identity providers can verify the software. Global Database can verify the company behind it.

How KYBA Could Fail

KYBA will not become credible enterprise infrastructure if it is reduced to a badge, a self-declared company name or an unexplained trust score.

KYBA should also avoid implying that verification guarantees good behaviour. A registered and financially healthy business can still breach a contract or misuse an agent.

Verification reduces uncertainty and establishes accountability. It does not eliminate commercial risk.

A Practical KYBA Adoption Roadmap

Most organisations do not need to build the complete future trust stack immediately. They do need foundations that avoid anonymous agents, unverified company claims and untraceable authority.

What KYBA Could Look Like Over the Next Five Years

Adoption is likely to begin in high-value workflows where enterprises already perform company verification: payments, supplier onboarding, financial services, insurance, credit risk and regulated data access.

Persistent agent identities

Commercial agents will require durable, revocable identifiers that can be recognised across platforms and transactions.

Machine-readable company credentials

Company registration, ownership, group relationships, authorised domains and source evidence will increasingly be delivered in standardised formats that policy engines can interpret.

Risk-based verification

A low-risk information request may require only agent authentication and active company status. A high-value transaction may require full ownership, financial, authority and evidence checks.

Dynamic access and pricing

A verified business identity could determine which data an agent may access, whether enterprise pricing applies, which payment terms are available and when human approval is required.

Mutual agent verification

In agent-to-agent commerce, buyer and supplier agents will need to verify each other's identities, businesses, authority and intended actions.

Continuous corporate monitoring

Agent behaviour and the underlying company will be monitored together. Changes in ownership, corporate status, financial condition or the authorising relationship could automatically alter the agent's permissions.

Every Commercial Agent Will Need a Verifiable Business Behind It

AI agents will become faster, more capable and more autonomous. That progress will increase the value they create, but it will also increase the consequences of unauthorised or incorrectly attributed actions.

Enterprises will need more than technical authentication. They will need to establish:

• Which agent acted.
• Who authorised it.
• Which legal entity is responsible.
• Who owns and controls that entity.
• What the agent is permitted to do.
• Which evidence supports the decision.

Know Your Business Agent provides a framework for connecting those elements.

It does not replace KYC, KYB, cybersecurity or delegated-authority controls. It binds them to a verified corporate identity so that an autonomous action can be traced back to the organisation responsible for it.

The market is already developing agent registries, agent credentials, transaction-intent protocols and edge enforcement. The remaining opportunity is to ensure that the company behind the agent is not treated as an unverified profile field.

In autonomous commerce, every agent will need an identity. Every commercial agent will also need a verified business behind it.