AI Sandboxing

The practice of running AI systems in isolated environments with restricted access to resources, data and external systems to contain potential harms.

Running AI in an isolated "sandbox" where it can't affect real systems or data. Like letting a child play in a fenced sandbox; they can experiment freely without the risks of the open playground.

Sandboxing is a governance control that reduces deployment risk. Your framework should require sandboxed testing for AI systems before they are granted access to production environments and real data.