How to solve AI's security problem | Anshu Sharma, Co-founder & CEO, Skyflow

Ashu Garg (0:00)

What's a common misconception people have about the security issues in AI?

Anshu Sharma (0:05)

The first misconception is the word security. Think about thousands of models and literally millions of agents running. None of the security, none of the privacy systems, none of our platforms were built around this idea that an arbitrary model or agent can do things that we can't predict because the agent and the model are non-deterministic. You can't actually isolate the security problems down to one thing. What you have to do is invert the problem and say, what do I need to do to have a secure privacy preserving, controllable infrastructure in place? At some point, you realize, you know what? There's probably a $100 billion company built just protecting the most sensitive data for two-person startups, all the way to Fortune 10 companies.

Ashu Garg (0:54)

We're here today with Anshu Sharma, the founder and CEO of Skyflow. Skyflow is an AI security company that is focused on securing sensitive customer data across data stores, models, and agents.

Welcome to the B2BACEO podcast, Anshu. It's good to have you today.

Anshu Sharma (1:13)

Great to be on the cast.

Ashu Garg (1:15)

Well, why don't you tell us a little bit about what Skyflow does and why it matters?

Anshu Sharma (1:19)

So yes, Skyflow is built on a very simple premise, which is that when we go about our day-to-day lives, we have no choice but to give our personal information to the governments we interact with. If you've tried filing taxes, you give them social security numbers, addresses. Even a coffee company, if you go to Starbucks, knows more about your location, has multiple credit and debit cards about you, knows actually where you travel, live half the time. As a result, we live in a society where constantly there's information being shared by the customers with these companies, and that forms the foundation of our society. Now, with AI, this just goes to the next level.

None of this will work if we can't trust the entities that we're sharing this information with. So how can I trust my personal information with the AI chatbot, with the hardware device that's in your ear, or the bank, if I can't be sure that that information is not going to end up in the wrong hands, inside their company, outside their company, or frankly be sold on the internet? So that's the problem we set out to solve, and we solve that with our platform.

Ashu Garg (2:30)

Makes sense. And that problem has always been true, as you rightly said, as long as there has been business relationships between consumers and a variety of enterprise, especially with the internet. But with agents, the problem has become 10x more. I was talking to one of my partners, and she was telling me how she's given instinct access to her bank accounts so that she could wire money to her landlord every month. And I was like, oh my God.

Anshu Sharma (2:56)

Every time we delegate an important task to another person, another business entity, or an agent, we're essentially trusting them with both information and workflow. And that cannot be done without deterministic controls. We live in a world where agents and models are not deterministic, which means they can arbitrarily do random things sometimes.

And if we want to use those things, we must have very, very deterministic, fine-grained controls in place. And none of the security, none of the privacy systems, none of our platforms were built around this idea that an arbitrary model or agent can do things that we can't predict. So we must somehow secure this information without the systems being deterministic. And that's a hard, fun problem to have, and that's what all of us are trying to build.

Ashu Garg (3:49)

And well, you've been working on this problem for many years. I mean, you started Skyflow now at this point, six, seven years ago.

You and I were brainstorming the idea for several years before that, and you worked on products that were sort of in the same general zip code for many years before that. Tell us a little bit about the founding story.

Anshu Sharma (4:09)

So if you actually think through the founding story, it has two or three points of pivot. One is, when I was at Salesforce, people don't remember this, but most of early Salesforce customers were actually small companies that trust the cloud. And my first problem was like, okay, how do I convince the JPMorgan, Chase and Ciscos of the world to give us their most important data? Well, turns out it's a trust problem. So we basically essentially had to rewrite a lot of Salesforce data stack, security stack, identity stack. We had to literally invent things like your ability to log into one application through another application. That's called an App Store or an App Exchange. We were the first company in the world to do that. So things like that had to be solved from first principles. And as I was solving them, I would go talk to these customers and like, hey, you're asking me to jump through all these hoops before you give me your data.