Capability Control

A mainstream approach in AI safety that constrains AI model behavior by restricting access to tools, channels, and external resources. It serves as a hard safeguard against uncontrollable outputs from AI agents.