# Sandbox

> A restricted environment where agents can run tools or code without affecting production systems.

- Category: Evaluation & Safety
- Stage: Established
- Also called: Execution sandbox, Isolation environment
- Canonical: https://www.joelcsaji.com/ai-dictionary/sandbox

## Plain English

A safe room for risky AI actions.

## Example

A coding agent runs generated code inside a disposable container with no production credentials.

## Why it matters

Sandboxes limit damage from mistakes, prompt injection, and malicious tool behavior.

## Related

- [Guardrails](https://www.joelcsaji.com/ai-dictionary/guardrails)
- [Action-Time Authorization](https://www.joelcsaji.com/ai-dictionary/action-time-authorization)
- [Computer Use](https://www.joelcsaji.com/ai-dictionary/computer-use)

## Sources

- [Anthropic: Building Effective Agents](https://www.anthropic.com/engineering/building-effective-agents)
- [OWASP Agentic AI Top 10](https://genai.owasp.org/2025/12/09/owasp-top-10-for-agentic-applications-the-benchmark-for-agentic-security-in-the-age-of-autonomous-ai/)