Urgent.News

What's breaking now, across thousands of outlets.

Tech

A sandbox you haven't tried to escape is a suggestion

Every agent incident report I've read this year ends the same way: someone assumed a boundary. The container "obviously" couldn't reach the network. The agent "obviously" didn't have the cloud keys. The socket "obviously" wasn't mounted. A sandbox is only a boundary where it removes capability. Everywhere else it's a naming convention with a Dockerfile attached. Docker's push into disposable…

Every agent incident report from this year ends with the same issue: a sandbox was assumed to have boundaries but failed to remove capabilities. Sandboxes are merely naming conventions with Dockerfiles attached, and Docker's push for disposable agent sandboxes is a step in the right direction. However, assuming a sandbox is always isolated is a mistake, as defaults are where most vulnerabilities occur.

The author provides a checklist to test agent containers for security. The checklist includes enumerating what the agent does not share with the host, such as the Docker socket, cloud credentials, SSH keys, and netrc. It also checks for read-write mounts, DNS resolution, TCP connections, and access to the instance metadata address.

The agent should run as a non-root user, and any writes should not persist after the container is removed. The author recommends running this checklist in CI on every image change to ensure the sandbox remains secure.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

More from Tuesday 17 November →