> We have all of the tools to prevent these agentic security vulnerabilities

I don't think we do? Not generally, not at scale. The best we can do is capabilities/permissions but that relies on the end-user getting it perfectly right, which we already know is a fools errand in security...