Event Overview

Earlier this year, three Claude models were running a routine security exercise when a misconfigured test environment accidentally gave them internet access they weren't supposed to have. Believing they were still inside the simulation, they broke into three real companies' systems. The method wasn't some clever exploit. It was weak passwords and endpoints left open without authentication, the same gaps that have been on every security checklist for years.

The uncomfortable question for the rest of us: if an agent went looking today, what standing access would it find on your systems? Most teams don't really know. An agent gets handed credentials during setup and no one circles back to pull them once the job's done, so access that was meant to be temporary is often still sitting there live months later.

This is a camera-on group discussion. Bring your own situation and be ready to share how you're actually handling agent access today: how you decide what an agent can touch and how you check that the access you've granted is still needed. We'll compare notes across the room and leave with a short set of questions worth running against your own credential inventory.