When the Agent Uses Your Login — Mandy Andress, CISO, Elastic

So let's start out by delving a little bit about your role at Elastic.

So from that vantage point, what can you see when it comes to AI and security that others may end up missing?

Yeah, absolutely.

I do feel like I sit in a relatively unique position, as you said.

CISO for Elastic, so responsible for securing Elastic infrastructure, enterprise, product security, customer trust, and then a component that we call Customer Zero, which is we are early and very aggressive users of our own products.

So putting it through its paces, a lot of people don't know the name Elastic, but it sits in the... core infrastructure of many, many, many of the online services and other companies that you interact with on a daily basis.

And the same goes for AI.

Elastic is a core component of a backend AI infrastructure.

So sitting at that convergence just allows us to see all the different ways that customers are trying to utilize AI, how, what's working, what's not, what types of data is interesting and relevant, and then how we can continue to analyze that deeper and understand within our environments how we need to adapt and adopt to AI.

So when you speak to security leaders and your customers, what do you think is the thing they most often get wrong when it comes to AI?

For me, these days, and things are changing so rapidly, I think they put a lot of emphasis on the underlying models.

And while that's critical and that's important, certainly models have their strengths and their weaknesses, and some are much better at different types of analysis and different types of applications.

But I see them not focusing as much on the data within their own organizations that they're utilizing within their AI processes.

It's making sure that that data is correct, is relevant.

Garbage in, garbage out, old saying still applies within AI.

And so focusing on the model versus focusing on what are the inputs to the models from your organization, you might get different answers if you had a different foundational data.