AI oversight at board level and detailed security information, but little disclosure of the models
Overall rating “solid”. No deduction for controversies. The superscript numbers link directly to the source.
1 · Responsibility
Visa documents board oversight, a data and AI governance function and mandatory training, but an AI policy, AI metrics and a complaints channel for those affected are missing.
What is documented
- According to the 2026 proxy, the Audit and Risk Committee oversees AI risks and the privacy programme; the full board reviews AI as part of strategy1
- A Data & AI Policy and Governance function under the Chief Data Officer and the Corporate Risk Committee manage AI risks2 3
- Policies on AI governance, model risk management and third parties exist; suppliers must apply responsible AI practices2 3
- The mandatory training “Responsible Data and AI Use” reached 100% completion in fiscal year 20252
- The 10-K names risks from AI and agentic commerce as risk factors6
The case against
Missing evidence of effectiveness
- No published AI policy, AI metrics or external review of AI management were found
- A complaints channel with documented remedies for those affected by AI-supported payment decisions is missing
- The Evident AI Index attributes operational implementation of responsible AI to other payment firms, not to Visa7
Sources for Responsibility (7)
- Visa Inc. (SEC EDGAR): Preliminary Proxy Statement for the 2026 annual meeting (v-20251126.htm) · 27 January 2026 · primary source
- Visa Inc.: 2025 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- Visa Inc.: 2024 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- Visa Inc. (SEC EDGAR, 10-K XBRL page): Form 10-K 2025, R33: Cybersecurity Risk Management and Strategy Disclosure (Item 1C) · 6 November 2025 · primary source
- Visa Inc.: Response to the Financial Stability Board's Consultation on Sound Practices for Responsible Adoption of Artificial Intelligence · 22 July 2026 · secondary source
- Visa Inc. (SEC EDGAR): Form 10-K for fiscal year 2025 (v-20250930.htm) · 6 November 2025 · primary source
- Evident Insights: Evident AI Index | Payments – Key Findings Report 2026 · 25 February 2026 · primary source
2 · Safety
Visa documents a comprehensive security programme, threat monitoring and agent recognition for agentic payments, but publishes no AI tests or error rates.
What is documented
- Item 1C describes a programme based on ISO/IEC 27002, PCI DSS and NIST CSF with around 1,000 specialists, three lines of defence and cyber fusion centres3
- The CISO reports twice a year to the Audit and Risk Committee; there were no material incidents in three fiscal years3
- A vulnerability disclosure programme via HackerOne with safe harbour is open to external researchers6
- The model inventory is monitored on a risk basis for suitability, performance and model drift1 2
- The Biannual Threats Report and a scam disruption unit document monitoring of AI-enabled fraud7
- For agentic payments, Visa describes Visa Intelligent Commerce in the 10-K and cryptographic agent recognition5 8
The case against
Gaps in AI testing
- Red-teaming of generative and agentic systems and independent model reviews are not documented
- Validation results and false decline rates of the fraud models are not published
- Visa aligns with standards; certifications such as ISO 27001 or SOC 2 are not named3
Sources for Safety (10)
- Visa Inc.: 2024 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- Visa Inc.: 2025 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- Visa Inc. (SEC EDGAR, 10-K XBRL page): Form 10-K 2025, R33: Cybersecurity Risk Management and Strategy Disclosure (Item 1C) · 6 November 2025 · primary source
- IT Brief Australia: Visa strengthens AI defences amid new era of cyber threats · 1 July 2026 · secondary source
- Visa Inc. (SEC EDGAR): Form 10-K for fiscal year 2025 (v-20250930.htm) · 6 November 2025 · primary source
- Visa Inc.: Visa Vulnerability Disclosure Programme · 3 November 2026 · primary source
- Visa Inc. (Payment Ecosystem Risk and Control): Biannual Threats Report Fall 2025 · 3 November 2026 · primary source
- Visa Inc.: Visa Introduces Trusted Agent Protocol: An Ecosystem-Led Framework for AI Commerce · 14 October 2025 · secondary source
- National Association of Attorneys General: State and Territory Attorneys General Urge Tech and Payment Platforms to Address Deepfake Exploitation · 26 August 2025 · primary source
- Visa Inc. (SEC EDGAR): Form 8-K on the voting results of the annual meeting of 2026-01-27 (v-20260127.htm) · 27 January 2026 · primary source
3 · Transparency
Visa describes AI products and risks in the 10-K and promises explainable models, but discloses neither model information nor limits or ways to challenge decisions.
What is documented
The case against
What is not disclosed
- No model or system cards, AI transparency report or training data information were found
- Explainability methods, error rates and limits of the fraud and authorisation models are not described3 4
- The EEA notices mention neither AI nor a right to human review of automated decisions2
- The number of models and investment in AI are known only from an interview5
Sources for Transparency (6)
- Visa Inc. (SEC EDGAR): Form 10-K for fiscal year 2025 (v-20250930.htm) · 6 November 2025 · primary source
- Visa Europe Limited: Global Privacy Notice – Jurisdictional Notice EEA, UK and Switzerland · 3 November 2026 · primary source
- Visa Inc.: 2024 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- Visa Inc.: 2025 Corporate Responsibility & Sustainability Report · 3 November 2026 · primary source
- IT Brief Australia: Visa strengthens AI defences amid new era of cyber threats · 1 July 2026 · secondary source
- Evident Insights: Evident AI Index | Payments – Key Findings Report 2026 · 25 February 2026 · primary source
4 · Why the harm rule does not apply
The harm rule does not apply: during the period there is no judgment, regulatory decision or fine concerning harm caused by Visa's own AI systems. The attorneys general's letter is not a decision, and the shareholder proposal was rejected.
5 · Controversies (2)
- minorThe 2026 annual meeting rejected a proposal for a report on online exploitation including AI-generated abuse content. · 27 January 2026 · abgewiesen
- minor47 attorneys general called on Visa and other platforms to stop payments for deepfake tools; not an enforcement action. · 26 August 2025 · unklar
6 · What remains open
- Item 3 of the 2025 10-K and the risk factors in the 10-Q for the June 2026 quarter
- The board's reasoning on the rejected shareholder proposal on AI-generated abuse content
- External audit or certification of AI management
- Model or system cards and the limits and error rates of the fraud and authorisation models
- Complaints channel and remedy for people affected by AI-supported payment declines
- Visa's response to the attorneys general on deepfake abuse content
7 · How it was assessed
Cut-off date: 3 November 2026 · 42 searches, 32 fetches.
Primary documents read (15)
- Visa Form 10-K 2025 (Item 1, Item 1A in part)
- Form 10-K 2025 Item 1C Cybersecurity (R33)
- Form 10-K 2025 Legal Matters (R30, R105)
- Preliminary Proxy Statement 2026 (PRE 14A), both parts
- Form 8-K Voting Results Annual General Meeting 2026-01-27
- Form 10-Q to 2026-06-30 (to Note 3) and Legal Matters (R25)
- 2024 Corporate Responsibility & Sustainability Report
- 2025 Corporate Responsibility & Sustainability Report (both parts)
- Visa's submission to the Financial Stability Board dated 2026-07-22
- Biannual Threats Report Fall 2025
- Global Privacy Notice EEA/UK/Switzerland
- Vulnerability Disclosure Programme
- DOJ notice dated 2024-09-24
- NAAG notice dated 2025-08-26
- SEC Rule 14a-8 letter dated 2025-09-17
Research of 3 November 2026 under Globalance AI Footprint rulebook version 0