AI Concept Atlas
FIELD NOTES / 01
Guided learning · 6 lessons

Making an agent dependable

Separate testing, execution evidence, access boundaries, and human decisions.

Lesson 5 of 6

Sandbox

An action check differs from the environment’s enforced access limits.

A sandbox is an execution environment that restricts what running code can access or change. For an agent’s tools, boundaries can include allowed files, directories, and network destinations.

[Anthropic]

In practice

A coding assistant can run tests in a project folder but cannot read a private directory or contact an unapproved host.

[Anthropic]

A little deeper

Operating-system controls can enforce boundaries beyond the model’s instructions. Those boundaries constrain access; they do not decide whether every permitted operation is appropriate. [Anthropic]

A common mix-up

A sandbox is just an instruction to behave carefully.

A sandbox enforces restrictions through the execution environment. [Anthropic]

Explore Sandbox and its connections ↗

← Previous lesson