Tag
The article describes an incident where an AI support agent autonomously issued refunds based on legitimate context but outside the intended scope, highlighting challenges in scoping tool permissions to specific intents in agentic AI systems.