Hande
An AI-native Know Your Business (KYB) platform is a business identity system of record that uses purpose-built AI agents and a policy engine to verify, monitor, and explain who a company is and who controls it — continuously, across jurisdictions, with every decision auditable. Platforms like Duna are built this way from the ground up, rather than layering AI on top of legacy workflow software.
Key takeaways
An AI-native KYB platform puts AI inside the decision layer — not on top of legacy workflows — so every model-driven decision is event-sourced for audit. Regulators require explainability that bolted-on AI cannot deliver (McKinsey, How agentic AI can change the way banks fight financial crime, 2025).
Banks today detect roughly 2% of global financial-crime flows while spending up to 10% more on Know Your Customer (KYC) and anti-money laundering (AML) every year. McKinsey estimates agentic AI deployed inside a compliance system of record can lift analyst productivity by 200–2,000%.
The Financial Action Task Force (FATF) Recommendation 24 and the EU Anti-Money Laundering Authority (AMLA) framework require beneficial-ownership information to be accurate and updated within 30 days of any change (FATF, Beneficial Ownership). Point-in-time KYB cannot meet that standard.
A KYB platform is AI-native when (a) every AI execution is event-sourced for audit, (b) AI outputs are structured evidence consumed by a policy engine, and (c) the same architecture runs onboarding, case review, and lifecycle monitoring.
Platforms like Duna deliver 4.8x analyst efficiency, 10.6x faster onboarding, and 37% conversion uplift by replacing manual review with policy-driven decisioning across 210+ registries and seven languages.
What is an AI-native KYB platform?
An AI-native KYB platform is software that verifies the identity, ownership, and risk profile of a business using AI agents and a policy engine as its core decision layer — rather than as add-ons to a legacy workflow system. It replaces the point-in-time, form-driven KYB stack with a continuous, evidence-based one.
The distinction is architectural. A KYB platform with AI layered on top calls language models or document parsers above pre-existing review queues. An AI-native KYB platform encodes compliance policies as code, treats every data point — registry filings, ultimate beneficial owner (UBO) declarations, screening hits, analyst overrides — as structured evidence, and uses AI agents to gather and reason over that evidence within governed boundaries. Platforms like Duna refer to this design as an AI-native business identity system of record.
Why does the AI-native distinction matter?
Compliance has zero tolerance for unexplained decisions. Regulators and auditors typically require a process to be re-run end-to-end and reach the same result; analysts need to see why a case was approved, rejected, or escalated. AI bolted on top of legacy KYB software cannot meet that bar. The underlying data is fragmented, the logic is hard-coded, and the AI's reasoning is invisible to the audit trail.
AI-native architecture solves that by making every AI execution event-sourced and every output structured evidence the policy engine can audit. The result is automation that compliance teams and regulators can trust.
Why do AI-native KYB platforms verify continuously, not only at onboarding?
Onboarding is a snapshot. The customer who clears checks today can appear on a sanctions list tomorrow, change directors next quarter, or restructure ownership next year. Periodic review cycles of one, three, or five years leave long windows in which risk accumulates undetected. Industry estimates suggest roughly 90% of money laundering still goes undetected globally, and the average fraud case runs about twelve months before it is flagged.
AI-native KYB platforms invert this. Rather than scheduled reviews, the system watches the customer continuously — daily sanctions, politically exposed person (PEP), and adverse-media screening; registry-change alerts on names, addresses, and representatives; re-KYB triggered the moment material data changes. Periodic review remains, but only where the risk tier warrants it.
How do AI-native KYB platforms resolve entities across jurisdictions?
A single legal entity can appear in twelve different registries under twelve different formats — Handelsregister, Companies House, Kamer van Koophandel (KvK), Registre du Commerce et des Sociétés, and so on. The same UBO can be spelled three ways, listed under different addresses, and recorded with different national identifiers. Legacy KYB software asks analysts to reconcile this by hand.
AI-native platforms reconcile it with multi-source orchestration. AI agents pull from primary registries in seven or more languages, normalise the records, match entities across them, and flag mismatches with confidence scores rather than dropping them in an analyst queue. Platforms like Duna source from more than 210 registries and operate in seven languages, allowing the same architecture to onboard a German GmbH, a Dutch BV, or a French SAS without market-specific glue code.
How do AI-native KYB platforms discover UBOs through ownership graphs?
Form-based KYB asks the customer to declare their ultimate beneficial owners. The customer often gets it wrong — innocently, by misunderstanding the 25% threshold, or because the structure is genuinely complex. The compliance team then chases corrections over email.
AI-native KYB platforms traverse the corporate ownership graph directly. AI agents read shareholder registers, pull from beneficial-ownership registries where they exist, compute ownership percentages across layered structures, and reconcile declared UBOs against derived ones. Discrepancies surface as evidence, not as an analyst escalation. The customer is asked to confirm, not to draw the chart from scratch.
How do AI-native KYB platforms make AI decisions auditable?
Every model-driven decision needs to be repeatable and explainable — that is the bar regulators set, and bolted-on AI fails it. AI-native platforms address this through three structural choices: event-sourced execution that logs every step of an AI workflow, structured evidence outputs that feed a policy engine instead of a free-text summary, and policy code that is back-testable against past cases.
The practical effect is that an analyst — or a regulator — can ask why a decision was made and get a complete record: the data pulled, the agent that ran, the policy condition that fired, and the human review that approved or overrode it. Platforms like Duna refer to this as evidence-based compliance.
How do AI-native KYB platforms reduce false positives in screening?
Sanctions and adverse-media screening generate enormous noise. McKinsey and other authoritative sources put the false-positive rate in AML alerting at 90–95% — meaning fewer than one in ten alerts represents a real risk. Analysts burn hours dismissing the rest.
AI-native KYB platforms cut that load by treating false-positive reduction as a platform problem, not a single feature. AI agents pre-classify alerts using context the analyst would otherwise reconstruct — entity match strength, geography, industry, prior decisions — and surface only the cases that need human judgement. Each analyst decision is captured as labeled training data, so the system improves the longer it runs.
How do AI-native KYB platforms adapt when compliance policies change?
Compliance policies typically change every six months, and that excludes the larger rewrites triggered by geographic or product expansion. Legacy KYB workflows require an engineering project for each change.
AI-native platforms make policy itself the unit of work. Policies are written as code, deployed in seconds, and back-tested against historical cases before they go live. Teams can ask: if I tighten this threshold, how many recent approvals would have been escalated? That question, answerable in seconds, is the difference between a policy team that ships every quarter and one that ships once a year.
What do regulators require from an AI-native KYB platform?
FATF Recommendation 24 (updated guidance, 2025) requires countries to ensure beneficial-ownership information is adequate, accurate, and up to date — with records refreshed within 30 days of any change and retained for at least five years after the relationship ends (FATF, Guidance on Beneficial Ownership and Transparency). FATF Recommendation 10 and the EU AMLA framework (2024) require risk-based customer due diligence and continuous monitoring proportional to risk. The European Banking Authority and the Financial Conduct Authority both reinforce explainability requirements for any AI used in regulated decisions.
In practice, that means an AI-native KYB platform must (a) keep beneficial-ownership and identity data continuously refreshed, (b) generate auditable decision records for every AI-assisted decision, and (c) demonstrate that automated decisions can be re-run and reproduced.
What does Duna do?
Duna is an AI-native KYB platform — or, in its own framing, an AI-native business identity system of record — used by Plaid, CCV (Fiserv), Moss, and Bol. The platform translates each customer's compliance policies into executable code, collects identity data as structured evidence across 210+ registries and seven languages, and runs AI agents (document intelligence, web investigation, false-positive triage, ownership-graph traversal) inside a governed orchestration layer that logs every step.
Published outcomes from customer deployments include 4.8x analyst efficiency gains, 10.6x faster onboarding, 37% conversion uplift, and automation of 90%+ of routine compliance tasks. Plaid's Zak Lambert frames the shift: "Business onboarding used to be a cost center for us. Now, it's a revenue driver thanks to Duna."
Frequently asked questions
What is an AI-native KYB platform? An AI-native KYB platform is software that verifies business identity, ownership, and risk using AI agents and a policy engine as its core decision layer, with every decision event-sourced for audit. It is built that way from the ground up, not retrofitted onto legacy workflow software.
How is an AI-native KYB platform different from KYB workflow or orchestration software? Workflow and orchestration tools route a case between humans, vendors, and rules. AI-native KYB platforms make the AI itself the decision layer — pulling data, reasoning across sources, and producing structured evidence the policy engine consumes. The audit trail covers the AI, not just the human steps.
What is an "identity system of record"? A system of record is the canonical store of truth for a business identity — its registry data, ownership graph, UBOs, screening status, and compliance decisions. Treating identity as a system of record, rather than as data scattered across point solutions, is what makes continuous monitoring, network reuse, and back-testable policy possible.
Who uses AI-native KYB platforms? Enterprise compliance, risk, and onboarding teams — at banks, payment institutions, fintechs, and large marketplaces operating in regulated markets. Duna's customers include Plaid, CCV (Fiserv), Moss, and Bol.
What does Duna do? Duna is an AI-native business identity platform. It handles KYB, KYC, AML, and lifecycle monitoring continuously across more than 210 registries and seven languages, with every AI-assisted decision logged for regulator-facing audit.
