How does Adyen use AI for fraud and risk?
Adyen's fraud and risk AI runs inside its own acquiring and banking stack. Protect, the risk module of Adyen Uplift that succeeded RevenueProtect, scores payments with machine learning and explains each decision; Dynamic Identification adds a behavioral trust model built from trillions of interactions; and Authenticate decides when a shopper needs to be challenged. Adyen's 2026 Fraud Report, published on May 21, 2026, says merchants using Protect saw fraud recall rise 16% and the false-positive rate fall 33% in 2025, and preserved twice as much authorization rate as merchants without it. Adyen also says Uplift cut false positives by 42% on average in 2025 and that pilot customers cut their manual risk rules by 86%. Fraud losses tied to fraudulent chargebacks on its platform fell 20% in 2025. Adyen applies the same identity layer to policy abuse: in H2 2025 pilots, a global luxury group found individual shoppers receiving up to €5,000 each in refunds. In compliance, Adyen says Dynamic Identification powers AI Investigator Agents for supervised, auditable reviews, without published figures. As a credit institution licensed by De Nederlandsche Bank, Adyen reports data-ethics risk in its annual report, has AI Ethics Principles and says it is preparing for the EU AI Act.[1][2][3][4][5][6][7][8]
Data as of Oct 19, 2026 · Adyen profile updated Oct 8, 2026 · 14 sources cited on this page · 52 fintechs profiled on this site · Markdown
Which fraud and risk models does Adyen run, and what does each do?
One row per model or product, with the latest result Adyen has published. Results are Adyen's own measurements.
| Model or product | What it does | Reported result | Date |
|---|---|---|---|
| Protect (Uplift risk module) | Machine-learning risk decisions in place of manual rules; a Fraud Signal Analysis shows which signal categories pushed a payment toward 'fraud' or 'legitimate' | Fraud recall up 16% and false-positive rate down 33% year on year; twice the authorization rate preserved compared with merchants without Protect | 2025[1][2] |
| Uplift risk pilot | AI-based risk management tuned to each merchant's risk appetite | Pilot customers cut manual risk rules by 86% on average, and 35% eliminated manual rules entirely | January 9, 2025[6] |
| Uplift at scale | Optimization and risk decisions across 6,500+ businesses | False positives down 42% on average in 2025; On's fraud rate fell 0.17% after moving to model-based risk assessments | 2025–2026[5][9] |
| Dynamic Identification | Behavioral trust model linking shoppers, businesses and credentials across Adyen's platform | Adyen says there is an 84% chance an identity has already appeared across its transactions, businesses, payouts or card issuing | November 2025 to May 2026[3][2] |
| Policy-abuse detection | Links refund activity at the identity level instead of per transaction | A global luxury group found shoppers receiving up to €5,000 in refunds, up to twenty times their average basket; a sports and entertainment customer found a shopper with about 70% of transactions refunded over several years | H2 2025 pilots[7][9] |
| Authenticate | Machine learning works with Protect to apply risk-based authentication, challenging only transactions that need it | Too Good To Go saw an 11% increase in payment conversion | September 22, 2026[4] |
| AI Investigator Agents | Process complex compliance reviews in a supervised and auditable way, using Dynamic Identification context | No figures published | November 11, 2025[3] |
| Agentic Payments | Fraud and authentication layer for agent-led transactions in Adyen Agentic | Limited availability for U.S. enterprise merchants; no results published | June 2026[10] |
What fraud results and data has Adyen published?
From Adyen's 2026 Fraud Report, which draws on US$1.6 trillion of 2025 platform data and a survey of 1,000 U.S. enterprise merchant decision makers, unless another source is cited.
| Measure | Figure | Period |
|---|---|---|
| Fraudulent-chargeback losses | Down 20% on Adyen's platform, while the average value of a fraudulent dispute fell 23% | 2025[2] |
| Risk approval rates in Japan, Australia and Singapore | Up to 99.57%, up an average of 17 basis points year on year, with chargeback rates falling in all three markets | 2025[2] |
| Legitimate customers blocked by static controls | Up to 10%, per Adyen platform data | 2025–2026[2][3] |
| Abuse types merchants report | First-party fraud 44.3%; fake accounts and identity abuse 42.2%; policy and promotion abuse 39.8% | Survey, 2026[2] |
| Costs of caution | 58% of merchants report rising manual review costs and 50% an increase in false declines | Survey, 2026[2] |
| Refund-policy tightening | Apparel and luxury merchants cut refund rates 21–25% without more disputes; elsewhere, a 38% cut in refund rates led to a 66% increase in chargeback volume | 2025[2] |
| Gift-card attack | A sportswear retailer's gift card activations surged more than 1,000% in August 2025; losses exceeded $750,000 before the attack was contained | August 2025[2] |
| Shopper recognition at peak | Nearly 95% of about 400 million unique shoppers recognized in real time | Black Friday and Cyber Monday 2025[11] |
How is Adyen's fraud AI governed, who carries the risk, and what is not disclosed?
Adyen is a licensed credit institution whose Management and Supervisory Board appointments need De Nederlandsche Bank approval. Its 2025 annual report defines data ethics risk as failing to meet internal principles and contractual and regulatory obligations on data, AI methods and models, and lists AI Ethics Principles, an AI and Data Ethics working group and preparation for the EU AI Act.[8]
The same report says Adyen observed accelerated use of AI-driven tools to scale financial fraud in 2025 and expanded its monitoring and threat intelligence; for financial crime it is using machine learning to bolster detection and preparing for the EU Anti-Money Laundering Regulation, which takes effect by July 2027.[8]
Chargeback risk sits with Adyen when it cannot recover disputed amounts from a merchant: the annual report describes this Merchant Potential Liability risk, a dedicated team and reserves withheld from merchant payouts. For agent purchases, Adyen says liability is unresolved between the merchant, the agent developer and the card issuer.[8][12]
What is not disclosed: absolute fraud rates on Adyen's platform, how many merchants use Protect, figures for AI Investigator Agents, and results for Agentic Payments. The Protect gains are year-on-year changes in Adyen's own metrics.[2][3][10]
What does the Adyen profile record on this topic?
The rows below are the entries of the Adyen AI strategy profile that bear on this page, as of Oct 8, 2026; the profile holds the full record of 14 dated moves, 7 use cases and 15 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Uplift Protect (risk module) | AI-based risk decisions replace manual rules; pilot customers cut manual risk rules by 86% on average, and in 2025 Uplift reduced false positives by 42% on average. On's fraud rate fell 0.17% after moving to model-based risk assessment. | In production | [6][5][9] |
| Adyen Agentic | APIs that translate one merchant integration across AI agent platforms and protocols, covering catalog feeds, cart building and agent-led payments with authentication and fraud controls. | Rolling out | [10][9] |
| Dynamic Identification for onboarding and compliance | A behavioural identity layer used to recognise known entities at onboarding and, Adyen says, to power AI Investigator Agents that process complex compliance reviews in a supervised and auditable way. | Announced | [3][13] |
| Behaviour-based underwriting | Adyen says Dynamic Identification lets it underwrite business financing using real commerce behaviour across its platform rather than forms and backward-looking files. | Announced | [3] |
| Rollout guardrails for code and ML models | Adyen says it is expanding investment in rollout strategies and rapid regression detection covering services, feature flags and ML models, and built Architect, a code, service and infrastructure graph available to humans and AI via MCP. | In production | [13] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Aug 13, 2026 | H1 2026: Uplift lifts conversion 0.9 points | Adyen reported net revenue of €1,302.9 million (up 19%) and processed volume of €803.8 billion (up 24%), said Uplift and Dynamic Identification increased customer conversion by 0.9 percentage points on average, and listed OpenAI among its enterprise wins. | [14][9] |
| Nov 11, 2025 | Dynamic Identification introduced | Adyen introduced Dynamic Identification, a behavioural model of trust built from trillions of interactions on its platform, used to strengthen Uplift, detect policy abuse, simplify onboarding and support AI Investigator Agents in compliance. | [3] |
What is Adyen Protect?
Protect is the risk module of Adyen Uplift and the successor to RevenueProtect, Adyen's risk engine. It makes machine-learning fraud decisions in place of manual rules. Adyen says merchants using it in 2025 saw fraud recall rise 16% and the false-positive rate fall 33% year on year.
Does Adyen's fraud AI explain its decisions?
Yes, per Adyen's documentation. The machine-learning engine returns a 'looks fraudulent' or 'looks legitimate' outcome with a Fraud Signal Analysis: a signal bar, signal categories and individual signal names showing what pushed the decision either way. Merchants can give feedback on each explanation.
How does Adyen detect refund and policy abuse?
With Dynamic Identification, which links refund activity to the same shopper across transactions and time. In H2 2025 pilots, a global luxury group found individual shoppers receiving up to €5,000 in refunds, and a sports and entertainment customer found a shopper who had refunded about 70% of purchases over several years.
How does Adyen handle fraud in agentic commerce?
Adyen Agentic includes an Agentic Payments layer for authentication and fraud in agent-led transactions, launched in June 2026. Adyen's September 2026 guidance names three new risks: agents abusing promotions and inventory, purchases that meet a mandate but miss the shopper's intent, and unclear liability.
Who supervises Adyen's use of AI in risk decisions?
Adyen N.V. is a credit institution licensed by De Nederlandsche Bank; its UK branch is authorised by the PRA and FCA and its San Francisco branch by the OCC and Federal Reserve. Adyen's 2025 annual report says it is preparing for the EU AI Act and has AI Ethics Principles.
- Explainable Machine Learning: the next phase of RevenueProtect (undated page, retrieved 2026-10-19)
- 2026 Fraud Report: Fraud's identity crisis
- Dynamic Identification: building the next layer of trust in global commerce
- Adyen Launches First-to-Market Upgrade to Authenticate, Unlocking Double-Digit Conversion Gains for Global Merchants
- Adyen Launches 'Personalize' to Tailor Checkout Experiences in Real-Time
- Introducing Adyen Uplift: The payment solution optimizing every transaction with AI
- H2 2025 Shareholder Letter
- Adyen 2025 Annual Report
- H1 2026 Shareholder Letter
- Adyen Announces Adyen Agentic As the Universal Translator for the Next Era of Commerce
- Adyen publishes H2 2025 financial results
- What is agentic commerce fraud, and how do you stop it?
- AI in our development lifecycle
- Adyen publishes H1 2026 financial results
When Adyen moves on AI, you'll read it here first.
when a bank, fintech or regulator moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning