Banking

“Knowing an Agent Is Deployed Isn’t Enough,” Says CrowdStrike’s Field CTO

01

Recognize that AI agents can behave unexpectedly in enterprise environments.

02

Implement stronger controls as agent capabilities become more advanced.

03

Understand that agents are designed to seek goals, potentially leading to unauthorized actions.

04

Be aware of prompt injection as a critical attack vector for manipulating agent behavior.

05

Monitor for risks like credential abuse and sandbox escapes in production settings.