Are we actually ready for AI-run cyberattacks?

We've spent years building security around the assumption that there is a human somewhere behind the keyboard.

But increasingly capable AI agents can perform multi-step tasks, interact with tools and operate with much less direct human intervention.

That's a very different threat model.

The scary part isn't necessarily "AI can hack everything."

It can't.

The more realistic concern is that AI can potentially make existing attack workflows faster, cheaper and easier to scale.

And the same technology can obviously help defenders too.

So I'm curious what people here think:

Does AI give attackers the bigger advantage, or will defenders ultimately benefit more from it?

And for organizations deploying AI agents:

Should an AI agent ever be given access to production systems without human approval for every sensitive action?

I'd genuinely like to hear the security community's perspective.

I'm especially interested in practical opinions from people working in SOC, pentesting, AppSec, cloud security and AI security.