AI oversight at board level, but the review procedure is described only in presentations
Overall rating “limited”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Mastercard documents board oversight and named persons responsible for AI, but describes its review and approval procedure only in conference slides and interviews, not in formal reports.
What is documented
- The 2026 proxy statement establishes ongoing board oversight of the ethical and responsible use of AI1
- The Chief Privacy, AI & Data Responsibility Officer briefs the Risk Committee, and the Chief AI & Data Officer the full board1
- Binding Corporate Rules are approved by the Belgian and UK data protection authorities; Mastercard was an early signatory of the EU AI Pact2 3
- The Impact Reports name the leadership of the AI governance programme and refer to AI standards in the supply chain and privacy impact assessments2 3
The case against
Review procedure not formally documented
- The inventory, scorecard and approval after control verification are described only in an employee's conference slides5
- Review before contracts are signed and an AI Governance Council are known only from interviews and a vendor case study6 7 8
- The AI policy and the risk classification scheme are not published
- No external audit or certification of AI management was found; the external audits mentioned concern cybersecurity and data protection2 3
Missing metrics and remedy
- Figures on reviewed, rejected or restricted use cases are missing
- No complaints channel with documented remedy is described for people affected by AI-supported decisions
Sources for Responsibility (10)
- Mastercard Incorporated (SEC EDGAR): Proxy Statement 2026 (Form DEF 14A, ma-20260427.htm) · 27 April 2026 · primary source
- Mastercard Incorporated: Mastercard Impact Report 2024 · July 2025 · primary source
- Mastercard Incorporated: Mastercard 2025 Impact Report · 15 July 2026 · primary source
- board-cybersecurity.com (reproduction of 10-K): Mastercard Inc 10-K Item 1C. Cybersecurity - 2026-02-11 · 11 February 2026 · primary source
- Mastercard Incorporated (Präsentation, DataCamp-Webinar): Four Pillars of AI Governance (slides, Alayna Kennedy) · 3 November 2026 · secondary source
- The Asian Banker: Mastercard embeds AI governance to support production-grade financial systems · 26 November 2025 · secondary source
- The CFO: How Mastercard's finance engine operationalized AI governance · 11 May 2026 · secondary source
- Credo AI: Mastercard and Credo AI: Scaling Generative AI Governance in the Financial Sector to Build Customer Trust (case study) · 3 November 2026 · secondary source
- Asia IP: LESI 2026: Mastercard executive warns of growing AI trust gap · 28 April 2026 · secondary source
- Mastercard Incorporated: Data Responsibility (company page) · 3 November 2026 · secondary source
2 · Safety
Mastercard documents cybersecurity with third-party reviews, alignment with standards and annual benchmarking, as well as model monitoring with bias testing, but publishes no validation or error rates for its models.
What is documented
- Item 1C describes CSO reports to the board, Risk Committee oversight, third-party reviews, ongoing testing and crisis simulations5
- Mastercard aligns with PCI DSS, ISO/IEC 27001 and SOC 1 and 2 and undergoes annual external benchmarking6 7
- A model monitoring API with automatic alerts and a bias testing API accompany the AI lifecycle1
- Agent Pay uses tokenised credentials and spending rules; Verifiable Intent is an open standard with tamper-proof authorisation8 9
- Mastercard had the fairness of its own models examined by a Data Study Group of the Alan Turing Institute4
The case against
Missing evidence of testing
- Red-teaming of generative and agentic systems and false alarm and decline rates of the fraud models are not published
- Performance claims for Decision Intelligence Pro come without methodology and are not independently documented3
- Item 1C was read only via an account, not in the original document5
- No usage restrictions are published for Recorded Future, which has government and military users11
Sources for Safety (13)
- Mastercard Incorporated (Präsentation, DataCamp-Webinar): Four Pillars of AI Governance (slides, Alayna Kennedy) · 3 November 2026 · secondary source
- The Asian Banker: Mastercard embeds AI governance to support production-grade financial systems · 26 November 2025 · secondary source
- CDO Magazine: Mastercard launches Decision Intelligence Pro, new generative AI model for fraud detection · 6 February 2024 · secondary source
- arXiv (Data Study Group, Alan Turing Institute): Measuring Fairness in Financial Transaction Machine Learning Models (arXiv:2501.10784) · 18 January 2025 · primary source
- board-cybersecurity.com (reproduction of 10-K): Mastercard Inc 10-K Item 1C. Cybersecurity - 2026-02-11 · 11 February 2026 · primary source
- Mastercard Incorporated: Mastercard Impact Report 2024 · July 2025 · primary source
- Mastercard Incorporated: Mastercard 2025 Impact Report · 15 July 2026 · primary source
- Mastercard Incorporated (SEC EDGAR): Form 10-K for fiscal year 2025 (ma-20251231.htm) · 11 February 2026 · primary source
- The Paypers: Mastercard introduces Verifiable Intent, co-developed with Google · 9 March 2026 · secondary source
- Asia IP: LESI 2026: Mastercard executive warns of growing AI trust gap · 28 April 2026 · secondary source
- The Register: Mastercard to buy Recorded Future for $2.65bn · 12 September 2024 · secondary source
- ThreatDown (Malwarebytes): Mastercard fixes potentially catastrophic DNS typo · 24 January 2025 · secondary source
- Information Age (Australian Computer Society): 'System update' triggers major Mastercard outage · 18 August 2026 · secondary source
3 · Transparency
Mastercard describes AI applications and principles in general terms and mentions internal model cards, but publishes neither system nor model cards nor an AI transparency report.
What is documented
- The 10-K and the Impact Reports describe AI applications in general terms, without system details1 3
- Mastercard commits to explaining how AI achieves its results and offers an opt-out for analytics2
- The Verifiable Intent specification is open and published with a reference implementation8
- Internal model cards are standardised; documentation templates are provided to bank clients6 7
The case against
System information not public
- Public system or model cards for Decision Intelligence, Brighterion and Agent Pay are missing
- The training data of Decision Intelligence Pro is known only as an order of magnitude from a media report5
- Limitations and error rates of the fraud models are not published; there is no AI transparency report
- No labelling of AI-supported decisions towards cardholders was found
Sources for Transparency (8)
- Mastercard Incorporated (SEC EDGAR): Form 10-K for fiscal year 2025 (ma-20251231.htm) · 11 February 2026 · primary source
- Mastercard Incorporated: Data Responsibility (company page) · 3 November 2026 · secondary source
- Mastercard Incorporated: Mastercard Impact Report 2024 · July 2025 · primary source
- Mastercard (Brighterion): Advanced AI technology datasheet · 3 November 2026 · secondary source
- CDO Magazine: Mastercard launches Decision Intelligence Pro, new generative AI model for fraud detection · 6 February 2024 · secondary source
- Mastercard Incorporated (Präsentation, DataCamp-Webinar): Four Pillars of AI Governance (slides, Alayna Kennedy) · 3 November 2026 · secondary source
- The CFO: How Mastercard's finance engine operationalized AI governance · 11 May 2026 · secondary source
- The Paypers: Mastercard introduces Verifiable Intent, co-developed with Google · 9 March 2026 · secondary source
4 · Why the harm rule does not apply
The harm rule does not apply: in the review period there is neither a judgment nor an authority decision nor a fine concerning harm from its own AI. The fine against Vocalink concerns risk management without an AI reference and does not count as a controversy.
5 · Controversies (5)
- minorA planned system update caused an outage of around 90 minutes in August 2026 with declined transactions, mainly in Australia; no AI reference. · 15 August 2026 · abgeschlossen
- minorA security researcher uncovered a typo in a Mastercard nameserver record that had persisted for years; no damage is documented, no AI reference. · 24 January 2025 · abgeschlossen
- minorThrough Recorded Future, Mastercard offers AI-supported threat intelligence that, according to its CEO, is also used by intelligence services and armed forces; no allegation of misuse was found. · 12 September 2024 · unklar
- minorThe FTC required Mastercard and seven other firms to provide information on AI-supported individualised pricing (“surveillance pricing”); this is a study without findings. · 23 July 2024 · unklar
- minorSeveral shareholder proposals were rejected at the 2024 and 2025 annual meetings, including a report on privacy and human rights; none concerned AI. · 18 June 2024 · abgeschlossen
6 · What remains open
- External audit or certification of AI management
- Public model or system cards and limitations of the fraud models
- Red-teaming and misuse monitoring for generative and agentic systems
- Complaints channel and remedy for people affected by AI-supported payment declines
- Full list of legal proceedings from Note 19 of the 10-K, and Item 1C in the original document
7 · How it was assessed
Cut-off date: 3 November 2026 · 44 searches, 32 fetches.
Primary documents read (11)
- Mastercard Form 10-K 2025 (Item 1 and Item 1A to p. 29)
- Item 1C Cybersecurity 10-K 2025 (reproduction)
- Proxy Statement 2026 (DEF 14A), both parts
- Form 8-K Annual General Meetings 2024 and 2025
- SEC Rule 14a-8 Response dated 2025-04-03
- Mastercard Impact Report 2024 (full text)
- Mastercard 2025 Impact Report (full text; data tables not extracted)
- Form 10-Q Q2 2026 (to Note 6)
- FTC notice dated 2024-07-23
- Corporate page: Data Responsibility
- Mastercard presentation: «Four Pillars of AI Governance»
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0