How does Santander govern AI responsibly?
Santander governs AI through a Data and Artificial Intelligence Corporate Framework approved by its board, which sets five principles for every use case: transparency (users always know when they are dealing with AI), fairness and bias mitigation, accountability with named owners across the lifecycle, privacy and security, and contribution to value and well-being. Oversight runs on a three-lines-of-defence model 'proportionate to risk', and responsible-AI training is mandatory for every employee working with these systems. In practice the group's Responsible AI team, led by Luis Ibáñez, tests conversational systems for over-refusal and unreliable answers, and a bank-wide evaluation framework described in July 2026 scores safety and fairness alongside accuracy before release and monitors products for drift afterwards; agents are built with stopping conditions, grounding, boundaries and audit logs. Santander says it does not share customer data externally to train third-party models, and it has open-sourced governance tools such as mech-gov-framework and Autoguardrails. In the US, Santander Holdings USA's 2025 10-K lists biased or flawed models, third-party model concentration and the cost of human oversight, testing and independent model validation as AI risks. No US-specific AI policy is published.[1][2][3][4][5][6][7]
Data as of Oct 16, 2026 · Santander profile updated Oct 13, 2026 · 15 sources cited on this page · 140 banks profiled on this site · Markdown
What are Santander's responsible AI principles and who oversees them?
From Santander's Data & AI page and group statements; they apply group-wide, including the US.
| Element | What it says | Source date |
|---|---|---|
| Data and AI Corporate Framework | Approved by the board; sets the principles every Data & AI use case is built and operated under | Retrieved October 2026[1] |
| 1. Transparency | Users always know when they are interacting with AI | Retrieved October 2026[1] |
| 2. Fairness and bias mitigation | Avoid unfairly discriminatory outcomes; mitigation proportional to risk and applicable regulation | Retrieved October 2026[1] |
| 3. Accountability | Comply with standards and regulation across design, implementation and monitoring; clearly defined owners | Retrieved October 2026[1] |
| 4. Privacy, data protection and security | Data minimization, anonymization, privacy by design, impact assessments | Retrieved October 2026[1] |
| 5. Value and well-being | Clear value for customers, employees and the organization, avoiding uses with disproportionate social or political impact | Retrieved October 2026[1] |
| Oversight model | Three lines of defence with independent oversight proportionate to risk; mandatory responsible-AI training for employees working with AI systems | Retrieved October 2026[1] |
| Customer data | 'We do not share customer data externally to train third-party models'; model-risk management aligned with internal AI regulation | 2025-2026[5][8] |
How does Santander test and control its AI systems?
| Practice | What it does | Date |
|---|---|---|
| Over-refusal testing | Responsible AI team tests customer-facing conversational AI so it neither refuses legitimate requests nor answers unreliably | July 23, 2026[2] |
| Measurable AI products | One evaluation framework scores performance, behavior, safety and fairness; generates test cases from each product's requirements and Responsible AI obligations; continuous production monitoring for drift and vulnerabilities | July 30, 2026[3] |
| 'Loop' and harness engineering for agents | Explicit stopping conditions and iteration limits, grounding in verified sources, evaluation at each step, tool boundaries (read-only, reversible or human approval) and full observability for audit | July 27, 2026[4] |
| Open-source governance tools | mech-gov-framework for language-model governance mechanisms; gen-fraud-graph synthetic data with no real customer data; Autoguardrails for guardrail optimization | June-July 2026[6][4] |
| Governed agent knowledge (A2K) | Scoped authority, cited responses and audit trails when agents query enterprise knowledge bases | September 23, 2026[9] |
| Explainability and bias research | CSIC agreement covering explainability, federated learning, bias detection and mitigation and responsible governance | June 30, 2026[10] |
What AI risks does Santander Holdings USA disclose?
The 2025 10-K, filed March 2, 2026, says AI models 'may be flawed, trained on insufficient or poor-quality data, reflect unwanted forms of bias or contain other errors', that controls such as 'human oversight, testing and independent model validation' may raise costs, and that errors in AI used for control functions such as transaction monitoring or sanctions screening could cause compliance failures.[7]
It also flags dependence on third-party models, datasets and infrastructure and 'concentration in a limited number of providers', data leakage into third-party training sets, and AI-enabled attacks including deepfake-enabled fraud. The Q1 and Q2 2026 10-Qs do not mention AI.[7][11][12]
The Webster acquisition needed approval from the OCC, the ECB and the Federal Reserve, which gave the last of the three on August 4, 2026; those supervisors' model-risk and AI expectations reach the US bank, the group and the US holding company respectively.[13]
What does the Santander profile record on this topic?
The rows below are the entries of the Santander AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 34 dated moves, 15 use cases and 47 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Machine-learning auto credit risk on FICO Platform | Approve/decline decisions, adverse-action reasons and probability-of-default for pricing in the US auto business, with faster model development and governance. | In production | [14] |
| Deepfake-scam education and AI fraud models | Purpose-built fake AI ads in the UK; consumer education in the US. | In production | [15] |
| Responsible-AI testing of conversational systems (group) | Santander tests customer-facing conversational AI for over-refusal and unreliable answers; the group also publishes an evaluation approach for AI products. | Rolling out | [2] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Jul 31, 2026 | Q2 10-Q filed, again with no AI language | Santander Holdings USA's second-quarter 2026 10-Q describes the Webster merger and SBNA's Openbank platform but has no AI discussion. | [12] |
| Jun 25, 2026 | Santander AI Lab publishes 11 open-source projects | Apache 2.0 repositories including gen-fraud-graph (synthetic fraud networks) and mech-gov-framework (language-model governance mechanisms); no real customer data. | [6] |
| May 4, 2026 | Q1 10-Q filed with no AI language | Santander Holdings USA's first-quarter 2026 10-Q does not mention AI; the 2025 10-K had carried an AI risk factor. | [11] |
| Mar 2, 2026 | 10-K names AI as a risk factor | Santander Holdings USA says it uses AI to improve customer experience and reduce operational risk, and warns of flawed or biased models and misleading third-party AI outputs. | [7] |
| Dec 1, 2025 | Deepfake ads made on purpose | The UK unit creates fake AI ads to teach customers what to watch for; 74% could not spot one. | [15] |
What are Santander's responsible AI principles?
Five, set in a Data and Artificial Intelligence Corporate Framework approved by the board: transparency, so users know when they are dealing with AI; fairness and bias mitigation; accountability with named owners across the lifecycle; privacy, data protection and security; and contribution to value and well-being. Oversight follows a three-lines-of-defence model proportionate to risk.
Who leads responsible AI at Santander?
Luis Ibáñez is Head of Responsible AI for the group; Ricardo Martín Manjón is Chief Data & AI Officer and Iñaki Bernal Chief AI Officer. Santander AI Lab, led by José Manuel de la Chica, publishes governance tools. No US-specific responsible AI lead is named.
How does Santander test its AI before customers use it?
Its Responsible AI team stress-tests conversational and agent systems for bias, unsafe answers, over-refusal and multi-turn manipulation, and a bank-wide framework described in July 2026 scores safety and fairness alongside accuracy before release and keeps monitoring products in production for drift and new vulnerabilities.
Does Santander use customer data to train AI models?
Santander says it does not share customer data externally to train third-party models and that AI-enabled processes run in secure environments. Its open-source tools, such as gen-fraud-graph, use synthetic data with no real customer data.
What AI risks does Santander US report to the SEC?
Santander Holdings USA's 2025 10-K warns that AI models may be flawed or biased, that third-party AI may produce misleading outputs, that it depends on a limited number of model providers, that controls like human oversight and independent validation add cost, and that AI enables cyberattacks such as deepfake fraud. The 2026 10-Qs do not mention AI.
- Data & Artificial Intelligence at Santander: Our Responsible AI principles (undated page, retrieved 2026-10-16)
- Designing conversational AI that knows when to say no
- When speed meets quality: How we make AI products measurable at Santander
- Engineering the loop: how Santander makes AI agents reliable
- Santander turns its AI-first strategy into measurable impact and extends AI access to all 185,000 employees
- Santander publishes AI projects under an Open Source licence to ramp up shared innovation
- Santander Holdings USA, Inc. Annual Report on Form 10-K for 2025
- Santander's data & AI-first strategy accelerates through OpenAI collaboration
- A2K: a new protocol suite for connecting AI agents to governed, auditable enterprise knowledge
- Santander and the CSIC join forces to spearhead research into advanced and responsible AI for the future of banking
- Santander Holdings USA, Inc. Quarterly Report on Form 10-Q for the period ended 31 March 2026
- Santander Holdings USA, Inc. Quarterly Report on Form 10-Q for the period ended 30 June 2026
- Fed gives final approval needed for Santander-Webster merger
- Santander US Auto Business Wins 2025 FICO Decisions Award for Machine Learning Credit Risk Analysis
- Santander creates fake AI ads for shopping scam awareness
When Santander moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning