How does BBVA govern AI and practice responsible AI?
BBVA published eight responsible-AI principles on July 8, 2026: transparency, fairness, precision and accuracy, privacy, data quality, technological security, responsibility and training, and social and environmental commitment, applied through technology, governance mechanisms and training and aligned with the EU AI Act. Below the principles sit working controls. Every assistant built on BBVA Assistants, the corporate generative AI platform, must pass the AI Assistant Governance Framework, which BBVA describes as mandatory and which sets human oversight, reviews of prompt quality and knowledge sources, and impact assessment. For classic models, the ADA platform's MLOps workflow blocks data scientists from approving their own models, promotes by model risk tier and keeps a central audit trail. For agents, 'The Frame', presented on October 6, 2026, gives each agent defined access, operational limits and human-oversight points, and in October 2026 BBVA began exploring NVIDIA's OpenShell architecture for agent security. Staff get a mandatory 'AI Express' course before Gemini access. Research backs it: Fair Learning, a three-year bias project with the University of Navarra started in August 2025. The 2025 Form 20-F names AI as a cyber risk.[1][2][3][4][5][6][7][8]
Data as of Nov 17, 2026 · BBVA profile updated Nov 15, 2026 · 21 sources cited on this page · 141 banks profiled on this site · Markdown
Which AI controls has BBVA described, and where do they apply?
| Control | What it requires | Applies to | Date |
|---|---|---|---|
| Eight responsible AI principles | Transparency, fairness, precision and accuracy, privacy, data quality, technological security, responsibility and training, social and environmental commitment | All AI at BBVA | July 8, 2026[1] |
| AI Assistant Governance Framework | Mandatory for all AI assistants developed within the bank: human oversight mechanisms, reviews of prompts and knowledge sources, impact assessment | Assistants on BBVA Assistants | July 31, 2026[2] |
| Model governance framework (ADA) | No data scientist can approve their own model; promotion by model risk tier; centralized audit trail of every lifecycle event | Machine-learning models on ADA | July 23, 2026[3] |
| The Frame | Defined access to systems and data, limits on autonomous action, human-oversight points, performance monitoring | AI agents | October 6, 2026[4] |
| OpenShell Reference Architecture | Exploring new approaches to AI agent safety, security and governance with NVIDIA | AI agents | October 21, 2026[5] |
| AI Express course | Mandatory training aligned with the EU AI Act before access to Gemini tools | Employees | July 2, 2025[6] |
| Golden datasets | A system that monitors every agent operation and response against benchmark datasets, in development | AI agents | November 11, 2026[9] |
Who works on responsible AI at BBVA, and what risks does it disclose?
Fair Learning, coordinated by BBVA's Data University and the University of Navarra's DATAI, brings twenty experts together for three years to detect, mitigate and correct AI bias and define best practices around the EU AI Act. Josep Amorós, the BBVA coordinator, said AI development must adhere 'to principles of fairness, responsibility and transparency'.[7][10]
Explainability is also a research line: Jesús Renero's ReX method, from the BBVA and University of Navarra industrial PhD program, aims to make AI more transparent in lending and risk management. Clara Higuera is BBVA's Responsible AI Lead.[11][12][13]
In its 2025 Form 20-F, BBVA says the 'prolific use' of AI has increased the risk of unauthorised access to its IT systems and client accounts. Not disclosed: a named board committee for AI, or counts of assistants rejected under the governance framework.[8]
What does the BBVA profile record on this topic?
The rows below are the entries of the BBVA AI strategy profile that bear on this page, as of Nov 15, 2026; the profile holds the full record of 45 dated moves, 19 use cases and 65 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Google Workspace with Gemini, Gemini app and NotebookLM | Offered to the whole workforce from July 2025 once employees complete the AI Express training. | In production | [6][14] |
| MLOps architecture with AWS inside ADA | Automates validation, traceability and control for models, with a centralised audit trail; pilots cut development time by 20% to 75% and infrastructure cost by 40% to 55%. | Rolling out | [15] |
| Responsible-AI framework and AI transition indicator | Eight principles implemented through technology, governance and training; the AI transition indicator reviews every client's vulnerability to AI disruption. | In production | [1][16] |
| 'The Frame' agent framework | Common standard to create, deploy, manage and govern agents: defined system and data access, action limits, human-oversight points, performance monitoring and cost and business-impact measurement. | Rolling out | [4][17] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Oct 29, 2026 | Q3 2026 results: AI-driven transformation and The Frame | BBVA's results release says the bank is progressing on its strategic plan with a particular focus on AI-driven transformation, and that The Frame will accelerate it; per MarketBeat on Yahoo Finance, management said a voice assistant reached close to a 90% successful-resolution rate in pilot interactions and some AI agents cut preparation time for certain enterprise risk financial programmes by as much as 30%. | [18][19] |
| Oct 21, 2026 | NVIDIA OpenShell collaboration on AI agent security | BBVA says it will work with NVIDIA to explore the OpenShell Reference Architecture for AI agent safety, security and governance, as part of scaling agents across the organisation; Global CSO Valentín Sánchez says expanding agents makes their security a priority. | [5] |
| Oct 6, 2026 | Strategic Talks #2: 'The Frame' presented to investors | Torres Vila, Bravo and Casas present the agent framework; BBVA reports assistants filtering over half of relationship-manager conversations in Spain, voice agents resolving close to 90% of interactions in pilots in Peru and Mexico, developer productivity up about 28%, and plans to roll Blue out to customers gradually in 2027. | [4] |
| Jul 30, 2026 | Q2 results release names 'The Frame' | The release says the bank has eight AI initiatives and a global framework, 'The Frame', to industrialise the creation, deployment, management and governance of AI agents at scale; on the call Genç adds that more than 100,000 colleagues actively use AI. | [17][16] |
| Jul 8, 2026 | Responsible-AI framework of eight principles | Transparency, fairness, precision and accuracy, privacy, data quality, technological security, responsibility and training, and social and environmental commitment, applied through technology, governance mechanisms and training and aligned with the EU AI Act. | [1] |
| Jun 4, 2026 | CEO at Money20/20: scaling agents is the priority | Genç reported 50% time savings in software development and testing and a 78% reduction in the time needed to record user complaints, and said all AI creation platforms will carry a cross-cutting security and responsible-use layer. | [20] |
| Feb 20, 2026 | 2025 Form 20-F lists AI as competitive and cyber risk | New technologies including artificial intelligence could erode the Group's business or require substantial investment; the 'prolific use' of AI has increased the risk of unauthorised access to BBVA's IT systems and client accounts. | [8] |
| Jul 2, 2025 | Gemini for the whole workforce | Google Workspace with Gemini, the Gemini app and NotebookLM for the bank's over 100,000 employees, with a mandatory 'AI Express' course aligned with the EU AI Act; access is granted after training. | [6][21] |
What are BBVA's responsible AI principles?
BBVA published eight principles on July 8, 2026: transparency, fairness, precision and accuracy, privacy, data quality, technological security, responsibility and training, and social and environmental commitment. They are applied through technology, governance mechanisms and training and aligned with the EU AI Act.
How does BBVA control its generative AI assistants?
Every assistant built within the bank must pass BBVA's AI Assistant Governance Framework, a mandatory internal model that sets human oversight, reviews of the quality of prompts and knowledge sources, and impact assessment. The Italy and Germany contact-center copilots were assessed under it before production.
How does BBVA govern AI agents?
Through The Frame, presented on October 6, 2026: each agent gets defined access to systems and data, limits on what it can do alone, human-oversight points and performance monitoring. BBVA is also exploring NVIDIA's OpenShell architecture for agent security and building monitoring against golden datasets.
Does BBVA research AI bias?
Yes. Fair Learning, a three-year project started in August 2025 with the University of Navarra, has twenty experts working on detecting, mitigating and correcting AI bias. A joint industrial PhD program also produced ReX, an explainable AI method.
- BBVA's Eight Responsible AI Principles
- BBVA Cuts Customer Inquiry Response Times by More Than 15% in Italy and Germany Thanks to AI
- MLOps transformation at BBVA: from data platform to scalable ML
- BBVA creates a model to scale AI across the Group and drive its customer-centric transformation
- BBVA explores new approaches to AI agent security with NVIDIA
- BBVA deepens partnership with Google Cloud to innovate with AI
- BBVA and the University of Navarra embark on a ground-breaking applied research project into responsible AI
- Form 20-F annual report for the fiscal year ended December 31, 2025
- Antonio Bravo (BBVA): “AI Will Help Us Build a Better, More Profitable Bank With Greater Growth Potential”
- BBVA and the University of Navarra embark on applied research project into responsible AI
- BBVA Completes Its First Industrial Doctorates in Data Science in Partnership with the University of Navarra
- Causality in AI: From Predicting What Happens to Understanding Why It Happens
- Responsible AI Is Not Just About Regulation—It Starts in the Design
- BBVA drives AI adoption through talent
- BBVA develops a new technology architecture with AWS to accelerate its AI solutions
- Banco Bilbao Vizcaya Argentaria (BBVA) Q2 FY2026 earnings call transcript
- BBVA Posts Over €6 Billion in Profit for the First Half of 2026 (Up 11 Percent) and Announces a New Share Buyback
- Earnings: BBVA Exceeds €9 Billion in Profit in the First Nine Months of 2026 and Increases Lending by 20 Percent Year-on-Year
- What Banco Bilbao Vizcaya Argentaria (BBVA) Said on Its Q3 Earnings Call
- Onur Genç: "AI Will Transform Banking for the Better, the Key Lies in Scaling Its Benefits Across the Organization"
- How BBVA is using AI to save employees 3 hours a week
When BBVA 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 PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning