Explainable fraud models for merchants, but unpublished AI ethics principles
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Adyen cites AI ethics principles and a working group, but discloses neither the principles nor a Supervisory Board responsibility for AI nor an approval procedure.
What is documented
- The Audit and Risk Committee of the Supervisory Board oversees risk management and the internal control system1 2
- A Risk Committee of the Management Board with the CFO, CRCO and CTO manages risks1 2
- The legal team is responsible as the second line for privacy, AI ethics and legal risks1 2
- The annual report mentions AI Ethics Principles, an AI and Data Ethics working group and preparation for the EU AI Act1 2
- A general complaints procedure sets deadlines and allows escalation to the supervisory authority5
The case against
Unpublished governance
- The AI Ethics Principles are mentioned in the annual report but are not published1
- An AI-specific responsibility of the Supervisory Board, a risk classification and an approval procedure for purchased models are not documented1 2
- A job posting describes building a company-wide AI governance framework only as a plan4
Missing accountability
- AI-related metrics, training, certifications or external audits were not found
- A complaints procedure for consumers in the case of automated rejections is not documented5
- The governance chapter, the Report of the Supervisory Board and the 2025 sustainability statement could not be read
Sources for Responsibility (6)
- Adyen N.V. (Ablage im Lobbyregister des Deutschen Bundestags): Adyen Annual Report 2025 (printed version, compressed PDF) · 5 March 2026 · secondary source
- Adyen N.V. (filed via FinancialFilings): Adyen N.V. Annual Report 2025 (FY2025) · 5 March 2026 · secondary source
- Adyen N.V.: All resolutions adopted at Adyen's 2025 AGM · 15 May 2025 · secondary source
- Adyen N.V.: Job posting “Senior AI Governance Counsel” · 3 November 2026 · secondary source
- Adyen N.V.: How to make a complaint · 3 November 2026 · secondary source
- European Commission: AI Pact – Signatories of the pledges · 3 November 2026 · primary source
2 · Safety
Adyen documents a binding procedure for vulnerability reports and a detailed incident report, while model validation and independent review are documented only as principles.
What is documented
- The responsible disclosure procedure provides for deadlines of 72 hours, three working days and at most 60 days until remediation6
- Adyen published its own incident report on the DDoS attack of 21 April 2025 with a timeline and countermeasures7 8
- The annual report lists the attack as the only reportable incident in 2025, with no intrusion and no reportable data loss11 1
- Adyen cites a low risk appetite for model risk and A/B/n tests with control groups before new models go into production1 2
- According to the company, generative AI runs on self-hosted models with a human in the loop3 2
The case against
AI testing only as a principle
- Validation and independent review of critical models are documented only as principles; results were not found1
- Monitoring and performance metrics of the fraud models and red-teaming results are missing
Sources for Safety (11)
- Adyen N.V. (filed via FinancialFilings): Adyen N.V. Annual Report 2025 (FY2025) · 5 March 2026 · secondary source
- Adyen N.V. (Knowledge Hub): How Adyen does AI (Andreu Mora) · 16 August 2024 · secondary source
- Adyen N.V. (Knowledge Hub, Tech stories): Operational efficiency with LLMs (Hernández/Mora) · 27 November 2023 · secondary source
- Adyen N.V.: Privacy policy – Security · 3 November 2026 · secondary source
- Adyen N.V. (developer documentation): Security (Development resources) · 3 November 2026 · secondary source
- Adyen N.V.: Responsible disclosure policy · 3 November 2026 · primary source
- Adyen N.V. (Knowledge Hub): Mitigating a DDoS – April 2025 (incident report) · 21 April 2025 · primary source
- Techzine: Dutch payment company Adyen hit by three DDoS attacks · 22 April 2025 · secondary source
- Adyen N.V. (Knowledge Hub): Building a merchant-first foundation for agentic commerce · 15 October 2025 · secondary source
- The Paypers: Adyen launches Adyen Agentic as an integration layer for agentic commerce · 17 June 2026 · secondary source
- Adyen N.V. (Ablage im Lobbyregister des Deutschen Bundestags): Adyen Annual Report 2025 (printed version, compressed PDF) · 5 March 2026 · secondary source
3 · Transparency
The product documentation explains the signals and limits of the fraud models to merchants, but information on automated decisions is missing for affected consumers.
What is documented
- The risk results view shows merchants signal bars, signal categories and the decision boundary, with a feedback function on the explanations5
- The ML rules document risk levels, the most influential signals, limits of detection and how merchants control the threshold6
- The fraud privacy notice states that model outputs can lead to merchant decisions on access and authorisation2
- Adyen states in general terms that training data comes from global transactions and names the model families used7 8
The case against
Weak information for those affected
Missing model information
- Model or system cards and an AI transparency report were not found
- On training data there are only general volume figures, no information on the use of personal data7
Sources for Transparency (8)
- Adyen N.V. (via Nasdaq): Introducing Adyen Uplift: The payment solution optimizing every transaction with AI · 9 January 2025 · secondary source
- Adyen N.V.: Privacy policy – Fraud (Polish country version) · 3 November 2026 · secondary source
- Adyen N.V.: Adyen Privacy Statement (version of 2025-08-04) · 4 August 2025 · secondary source
- The Paypers: Adyen launches Adyen Agentic as an integration layer for agentic commerce · 17 June 2026 · secondary source
- Adyen N.V. (product documentation): Machine learning demystified (RevenueProtect, Risk results) · 3 November 2026 · primary source
- Adyen N.V. (product documentation): Machine learning rules (Configure your risk profile) · 3 November 2026 · primary source
- Adyen N.V. (filed via FinancialFilings): Adyen N.V. H1 2025 Shareholder Letter and Interim Condensed Consolidated Financial Statements · 14 August 2025 · secondary source
- Adyen N.V. (Knowledge Hub): How Adyen does AI (Andreu Mora) · 16 August 2024 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: within the 36-month window no judgment, regulatory decision, fine or pattern of serious incidents involving Adyen's own AI systems was found. The 2025 DDoS attack has no AI link.
5 · Controversies (1)
6 · What remains open
- Wording of the AI Ethics Principles and mandate of the AI and Data Ethics working group
- AI-specific responsibility of the Supervisory Board
- Risk classification of the AI systems under the EU AI Act and approval procedures for purchased models
- Results of model validation, independent review and monitoring of the fraud models
- Rights of affected consumers in the case of automated rejections
- Content of the 2025 sustainability statement (ESRS S4, G1)
7 · How it was assessed
Cut-off date: 3 November 2026 · 43 searches, 34 fetches.
Primary documents read (11)
- Adyen Annual Report 2025 (pp. 5–48, two filings)
- Adyen H1 2025 Shareholder Letter and interim financial statements (first 100,000 characters)
- Adyen H1 2026 Earnings Release
- Adyen Privacy Statement (version 2025-08-04), privacy notice on fraud and security
- Adyen Responsible Disclosure Policy
- Adyen product documentation on ML risk rules and Risk results
- Adyen incident report: DDoS attack, April 2025
- DNB Enforcement measures 2026
- Robeco note on the Adyen 2026 Annual General Meeting
- Court filing details: Hecker v. Adyen NV (N.D. Cal.) and UPC_CFI_0000662/2026
- European Commission, list of AI Pact signatories
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0