You'd be surprised to see how often we're seeing those types of semantic attack vulnerabilities in Agent frameworks: https://niyikiza.com/posts/map-territory/