Skip to content

Content Library

Back to Articles

Article December 19, 2025

AI Growth Forces Banks to Reinvent Model Validation and Governance

AI Growth Forces Banks to Reinvent Model Validation and Governance

Rita Gnutti, Executive Director Internal Validation and Controls, Group Chief Risk Officer Area, Intesa Sanpaolo LinkedIn profile for Rita Gnutti

Facing a surge in AI-driven models, Banca Intesa Sanpaolo is pioneering a transformation in model risk and validation with automation and GAI.
  • Banca Intesa Sanpaolo doubled models from under 300 to nearly 700 since 2020 with rising AI and ML adoption

  • Independent validation coverage fell below 70% driving framework transformation

  • New risk methodology scores inherent and residual risk adding AI factors like fairness explainability and hyperparameter robustness

  • Central inventory includes all AI systems proprietary and third party

  • Digital automation tool integrates methods testing and code reducing manual effort and backlog

  • Generative AI used in controlled closed knowledge base to assist code conversion testing and report drafting with human oversight

  • ML credit risk model framework evaluates overfitting interpretability and hyperparameter sensitivity balancing performance and risk

  • Regulatory focus including EU AI Act leads to tiered oversight for high risk AI and future expansion to compliance and ESG models

As artificial intelligence accelerates across financial services, institutions are rethinking how they manage and validate increasingly complex model ecosystems.


Rising model inventories, growing interconnectedness, and mounting regulatory expectations are pushing risk and validation teams into a new era in which traditional approaches no longer scale. 


And according to a leading model risk expert, the challenge now is building a validation function capable of safeguarding integrity while enabling innovation, even as the technological landscape shifts beneath it.
 

Speaking at CeFPro’s flagship Risk Evolve 2025 event, Rita Gnutti, Executive Director, Head of Internal Validation and Controls Head Office Department at Intesa Sanpaolo, described how her institution has undergone a wide-ranging transformation of its model risk and validation function.


She explained that the model landscape has grown dramatically since 2020, with the inventory rising from fewer than 300 models to nearly 700. 


This surge spans both risk management and business functions, introducing an entirely new generation of AI-driven and machine-learning-based models. 


According to her, this expansion places mounting pressure on second line of defence teams, which must provide independent oversight despite resource constraints and rising model interconnectedness. 


As she noted, independent coverage has fallen below 70 percent, making transformation essential.


Gnutti stated that her team has responded by redesigning the model risk management framework, and she highlighted a structured assessment methodology that evaluates inherent and residual risk across model design, implementation, data quality, performance, and use. 


She emphasized that AI-specific dimensions such as explainability, fairness, hyperparameter robustness, and transparency are now integrated directly into risk assessments. 


Gnutti also explained that all AI systems, whether formally classified as models or not, are centrally inventoried to ensure oversight across both proprietary and third-party applications.


A second stage of transformation focused on digitalization. Gnutti described how her team built an automation tool to streamline ongoing validation, integrating methodological handbooks, testing procedures, and validation code into a unified digital workflow. 


This tool, she said, significantly reduces manual workload and allows experts to concentrate on higher-value analysis. Early use demonstrated substantial time savings, strengthening the case for broader automation.


She added that the team is now piloting generative AI within validation activities and explained that generative AI is being tested for tasks such as converting legacy code, interpreting model test results, and drafting standard validation reports. 


This is done within a tightly controlled environment supported by a closed knowledge base of internal documentation and always under human supervision. 


She stressed that this human-in-the-loop approach preserves interpretability while exploring efficiency gains. Early pilots show promise, but she emphasized the need for careful governance to avoid overreliance.


Gnutti also presented a machine-learning-specific validation framework, noting that traditional principles still apply but must be expanded. 

Events you may be interested in

View all of our upcoming events

Sign Up to Our Newsletter

Sign up

Connect Plus

Unlock more with Connect Plus membership

This content is available to Connect Plus members.