The attacker needs sustained, useful output. Defenders can turn that dependency into friction across the access ecosystem.

Everyone wants the AI bug hunter.

Fewer people want the patch clock that comes with it.

That’s the part getting buried under the stage fog: if the models are better at finding and understanding real bugs, your org does not get safer by applause. It gets safer if it can move before somebody else does.

Anthropic just showed what happens when your “helpful” AI agents become C2: 80–90% of an espionage op automated, humans just clicking approve.

Lock down identity + connectors or you’re renting your SaaS to someone else’s botnet. 🤖🚨