Responsible AI in the enterprise – at Legal & General

Top three takeaways

• What sets generative AI apart from earlier enterprise technologies is its ability to generate new data, raising fresh questions about how models operate, learn, and avoid bias.
• Legal & General is not starting from zero on governance — years of managing machine learning have given it strong model risk policies and regulatory-compliance frameworks.
• Existing frameworks need to be augmented rather than rebuilt, with particular attention to model validation and the ethical implications of generative capabilities.
• Ensuring models remain compliant, meet regulatory expectations, and treat customers fairly stays central as governance evolves one step further to meet the new technology.

Stefana Brown — Group Technology and Data Risk Director, Legal & General[00:10]

Hi, my name is Stefana Brown. I’m the Group Technology and Data Risk Director at Legal & General.

David Cushman — HFS Research[00:10]

Thank you for joining us today. We’ve been talking a lot about generative AI for the last 18 months, two years I guess. We’ve had new technologies come in and make an impact on the enterprise before. How is this one different in terms of how we have to manage our governance around it?

Stefana Brown — Group Technology and Data Risk Director, Legal & General[00:34]

The main difference with generative AI to other previous technologies is really its ability to generate new data, which is not what we have had with previous technologies. Therefore, our understanding of how our models operate and how they learn and avoiding bias in that space is really important. It’s something that we haven’t had to necessarily deal with before.

David Cushman — HFS Research[00:58]

Your previous governance processes, were they set up to handle that? If not, how have they had to change?

Stefana Brown — Group Technology and Data Risk Director, Legal & General[00:58]

Our governance — we are not starting from zero. We have had to deal with machine learning for a number of years. We have strong model risk policies. What we have always had to do is ensure that our machine learning models are compliant and meet our regulatory expectations and, of course, treat our customers correctly. So our frameworks are there. What they need to do is be slightly augmented to meet that generative element, so especially when it comes to validation, model validation, ethical implications, etc. So there are certain elements that we really need to look at in more detail, but in general, we are starting from a good place. We just need to move that one extra step further to make sure we can respond to this new development.

David Cushman — HFS Research[01:50]

Thank you very much.

Stefana Brown — Group Technology and Data Risk Director, Legal & General[01:50]

Thank you.