Generative AI: Balancing security with innovation

Learn how Discover is embracing GenerativeAI in a thoughtful, intentional, security-focused manner.
January 09, 2024

Approaching AI in a thoughtful, intentional, and security-focused manner is imperative for fintech companies to nullify potential threats and maintain customer trust while still taking advantage of AI’s power.

The speed at which artificial intelligence (AI)—and particularly generative AI (GenAI)—is upending everyday life and entire industries is staggering. Slowing the progression of AI may be impossible, but approaching AI in a thoughtful, intentional, and security-focused manner is imperative for fintech companies to nullify potential threats and maintain customer trust while still taking advantage of its power.

AI threats to fintech companies

When I think about possible AI threats, top of mind to me is how AI can be weaponized:

  1. Threats to identity. Whether it’s deep fakes or simply more sophisticated phishing attempts, AI is making it easier to steal identities and ramping up the need for more accurate, faster authentication.
  2. Misinformation and manipulation of data. As AI becomes more powerful, its ability to manipulate data is increasing and making it difficult to stem the tide of misinformation. Additionally, related issues during use are risk of hallucinations and prompt engineering.
  3. Exploiting technology vulnerabilities. Bad actors have the potential to train AI to spot and exploit vulnerabilities in tech stacks or business systems. While we can’t plan for every new threat that AI poses, it’s imperative to have the right AI usage guardrails in place at Discover® Financial Services and know how to quickly address any vulnerabilities.

Our approach to securing against AI threats and ensuring Responsible AI

At Discover, we've established an AI Governance Council, which consists of a cross-functional team of data scientists, cybersecurity experts, audit and compliance personnel, legal representatives, technologists, and decision-makers who collaborate to set standards to establish a framework for the adoption of AI in a responsible manner.

By including a wide range of participants who represent different facets of how AI is being used, unique use cases, and differing perspectives, we can create AI guardrails applicable across business units within Discover. Additionally, it’s paramount within the financial services sector to ensure responsible AI and adherence to regulatory guidance for model risk. Keeping our AI approach interpretable and managing bias becomes crucial.

At a high level, these guardrails relate to:

  1. Limiting access to all public large language models and preventing employees from using customer data within any public generative AI models
  2. Clear intake process that teams complete when they want to use public, vendor, or homegrown AI tools and models.
  3. Established risk management framework to evaluate the use cases and validate the controls to manage relevant risks
  4. Continuous authentication and authorization to maintain the principles of least privilege and context of user entitlement.
  5. Proper data labeling and logging to maintain confidentiality
  6. Human-in-the-loop validation to ensure each AI use case is reviewed and approved by a subject matter expert to ensure the accuracy and quality of the output is fit for purpose
  7. Recording inputs and documenting what we’re inputting into any language models—and the outcomes to ensure the integrity of the processes
  8. Established feedback loops so that we’re quickly getting and responding to feedback about using the models
  9. Required training for any employee using AI models in their work to ensure their work adheres to standards related to AI trust, transparency, trustworthiness, and the like.

As we deploy our guardrails, we also evangelize across teams at Discover through our internal learning platform, Discover Technology Academy, through various events and emails and required security training.

Managing GenAI testing and access with trusted partners

We don't have the luxury of waiting to see how AI evolves before it affects our everyday life. We must deal with the threats it poses in real time—while taking advantage of the competitive advantages it offers.

To us, that takes shape by using closed language models, with AI partners we trust, to run proof of concepts and other tests that help us understand how to use GenAI in a trustworthy and transparent way. We have partnerships with large tech companies to test their AI offerings and tools in controlled, managed experiments.


As the Chief Information Security Officer (CISO) at Discover, I am both excited and sober about how generative AI will change the fintech landscape in the coming years. The trust we build with our customers is our most important asset—and we don’t take that for granted. Having clear guidelines for how employees can engage with and use AI models and mechanisms to enforce guidelines will help us enable innovation while ensuring the security of our customers, their data, and their assets.

© 2023 Discover Financial Services. Opinions are those of the individual author. Unless noted otherwise in this post, Discover is not affiliated with, nor endorsed by, any of the companies mentioned. All trademarks and other intellectual property used or displayed are property of their respective owners