Archive
AI Security

The Acronym For The 2025 Defensive Paradigm

Archive
Hard
200pts0 solves
Google DeepMind's April 2025 paper proposed a design pattern in which a privileged LLM plans actions that are executed by a sandboxed, capability-restricted LLM — a step toward provable defense against prompt injection. Acronym?
Show hint
It is also an animal with a hump.

Archive — no submissions accepted

This challenge is preserved for reference. Play live challenges at /challenges.