Companies are rushing to integrate AI into workflows, but cybersecurity teams worry they're losing control of data. Microsoft Copilot can surface information across your organization that users shouldn't see, like salary data from someone else's OneDrive, because AI ignores traditional file permissions.
I think you see this rising at a very fast rate. You've got a lot of money going in the market, a lot of people developing software, and then beyond just developing these engines or LLMs, it's how fast can I integrate this into every part of my life, which is great. I don't want to discourage that. I think what happens is, is as cyber professionals look at integrating these tools, there's a lot of fear or reluctance because of you're losing control, right? I've got— I'm giving a system access to every bit and byte across my enterprise. And how do I know what it can see, what it can't see? And so let's talk about Copilot, for example, with Microsoft. That's probably a very popular one people may, may know about here. AI is not bound to the same permissions like a file server would be bound to, right? Like Bronwyn's files are Bronwyn's files, Cody's files are Cody's files, and I can't see Bronwyn's and Bronwyn's can't see mine. Now we put in Copilot into the system and say, I prompt it and say, looking for information on salaries for this area. If there is a document in your OneDrive that is marked public, maybe I don't have, I can't see it right now. Copilot will see it and they can ingest that data and present it to the person that prompted.
