Content

AI coding agents can run commands, edit files, and explore your project on their own, which is exactly what makes them so powerful. Started directly on a personal device, a login or compute node; however, they inherit everything you have access to: your SSH keys, your entire project directory, shared data, and cluster resources. In this session, we show you why that’s a risk worth taking seriously and how containers and sandboxing help you decide what an agent can actually see and touch.

What you’ll take away

• What an AI agent can reach when it runs with your permissions on a shared cluster
• How containers and agent sandboxes (e.g., bubblewrap) limit access to files, keys, and network
• Simple patterns for giving an agent exactly what it needs for the task, and nothing more

Prerequisites

Basic command-line experience. No prior container knowledge needed.
 

Access

To access the event, click on the following link: 

Big Blue Button Link <https://webconf.hrz.uni-marburg.de/n/rooms/hsy-cmh-knb-ndw/join> 

Access Code: kw7i98