Ave Gatton
๐ค SpeakerAppearances Over Time
Podcast Appearances
We are currently seeing the rollout of agents in a lot of different industry verticals.
Finance is a large one.
There's also healthcare.
And then there's a whole slew of agents that are outward facing customer service.
agents.
Roughly speaking, I divide these categories into dividing these agents into two categories, one being outward facing to customers and one being inward facing where the customer is an internal employee of the company.
And you might expect that those internal facing agents are more secure because the users are a trusted individual.
But when we look at the recent trends in cyber security, you find that a lot of large companies now expect that bad actors will actually come from internal employees as opposed to external users.
And you have the added danger that the internal agents usually are given higher permissions
and have access to more sensitive data, specifically because they are being used by internal employees.
And of course, hearkening back to this concept that you can have an agent, if it doesn't have access to anything meaningful, then it can't really do anything useful.
You have to give it access to meaningful information in order for it to take meaningful actions.
So I would say that, in fact, the security becomes more important for internal facing agents that are talking to internal employees and are accessing more sensitive data than they are for, say, external facing public employees.
service agents or whatnot, what have you, that way you can pretty effectively lock down their ability to see sensitive information.
Yeah, it's a good question.
There's a lot to unpack there of like why you can't just take traditional security and apply it to an agent.
Certainly there's a lot of traditional security that you can take and apply to an agent.
Role-based access controls to certain, like to databases and other information.
But you have to think of an agent as taking on the capabilities of the user that's talking to the agent.
And in that case, there's multiple different paradigms.