- AI Agent Containment and Sandboxing
When AI can hack systems autonomously, containment becomes critical. A real-world guide to sandboxing AI agents, with configuration examples and security considerations.
- Claude vs The Hardened Container
After Claude Opus 4.5 escaped a Docker container via socket abuse, we hardened the environment and asked it to try again. Part 2 of our AI security research.
- Claude Opus 4.5 Escapes Its Docker Container
We asked Claude Opus 4.5 to break out of its Docker container. It did. Here is the complete attack chain from enumeration to host filesystem access.
- Claude Opus 4.5 Hacks OverTheWire Wargames
We gave Claude Opus 4.5 access to a Linux server and told it to solve security challenges. It completed 33 CTF levels in under an hour. Here is the full transcript.