BNP Paribas Fortis has extended its collaboration with French artificial intelligence company Mistral AI for three years, marking a significant commitment to European technology infrastructure within the financial sector. The Belgian banking giant, which serves as the largest retail bank in the country, stated the decision was driven by data sovereignty requirements and the need to keep sensitive customer information within European jurisdiction.

The partnership, originally established in 2023, allows the bank to deploy Mistral's large language models in-house rather than relying on public cloud APIs from American providers. Manuel Piette, Chief Data Scientist and Head of AI at the bank, indicated that this architecture is non-negotiable for processes involving know your customer checks and risk management. The move aligns with broader regulatory expectations across the euro area regarding operational resilience and third-party risk.

While many financial institutions have experimented with generative AI, few have committed to a multi-year contract with a European model provider for core banking functions. The bank plans to focus the technology on high-value use cases in corporate and institutional banking, as well as commercial banking operations. This suggests a shift from experimental pilots to integrated workflow tools.

The sovereignty calculation

For Piette, the choice of Mistral over US hyperscalers is primarily about control. Banking data carries heavy compliance burdens, and moving such information outside the European Economic Area introduces legal friction. By hosting models on-premise or within trusted European clouds, the bank retains full expert validation of risks and decisions. This approach mitigates the risk of foreign legal mandates accessing customer data.

The bank described the partnership as part of a multi-model approach guided by performance and geographic considerations. This reflects a growing trend among European enterprises to diversify their technology stack to avoid vendor lock-in. The European Commission's digital strategy has long encouraged such sovereignty, though market reality often favours US incumbents. BNP Paribas Fortis is attempting to prove that viable alternatives exist for critical infrastructure.

Internal adoption and training

Technology procurement is only half the battle. The bank has established an AI Centre of Excellence to manage the cultural shift among its 11,000 employees. Piette notes that staff anxiety regarding job displacement is common. His team counters this by framing the tools as creativity enhancers rather than replacements. Journaling sessions are held every two weeks to demonstrate practical applications, such as evaluating AI agents or integrating voice-native technologies.

Training extends beyond data scientists to include business units that manage customer interactions. The goal is to attract talent that has already used AI in other environments while retaining institutional knowledge. Piette observes that the skill set required is evolving from pure statistics to software engineering and machine learning integration. This transition mirrors the broader labour market shift seen across the continent's technology sector.

The unstructured data challenge

A significant portion of the bank's data remains unstructured, described by Piette as the unseen half of an iceberg. Legacy documents, emails and handwritten notes resist traditional analysis. The bank is deploying AI to enhance internal document search and automate data extraction. This capability is critical for complex financial analyses that require cross-referencing multiple sources, a task that previously demanded substantial manual labour.

To manage this, the bank utilizes the Domino Data Lab platform to harmonize data across business units. Without this layer, models cannot access consistent inputs. The combination of harmonized data and sovereign AI models creates a closed loop where information flows securely from ingestion to insight. This architecture is now being tested against fraud detection and customer protection workflows.

Regulatory headwinds

Financial services remain one of the most heavily regulated industries in Europe. The European Central Bank and national supervisors scrutinize algorithmic decision-making, particularly where credit or risk is involved. Automating low value-added steps is permissible, but final decisions often require human validation. The bank insists that expert validation of risks and decisions will be maintained despite the increased automation.

Compliance requirements for know your customer processes continue to grow, adding complexity to any automation project. The three-year renewal suggests the bank is confident it can navigate these rules while achieving efficiency gains. If successful, this deployment could serve as a reference case for other European financial institutions weighing similar investments. Failure would likely reinforce reliance on established US providers.

People mentioned

  • Manuel Piette

    Chief Data Scientist and Head of AI, BNP Paribas Fortis

Organisations

BNP Paribas Fortis · Mistral AI · Domino Data Lab