It limits what untrusted or less-trusted activity can affect outside the boundary. Sandboxes protect browsers and applications, isolate tests, and support controlled observation of suspicious files or behavior.
Implementations use different mechanisms and trust assumptions, including operating-system process restrictions, application policies, containers, virtual machines, emulation, or combinations of controls. The meaningful question is not whether something is “sandboxed,” but what boundary is enforced, by which trusted components, and against which threats.
Key points
Policy designGrant only required capabilities, constrain communication and persistence, separate privileged brokers from restricted workloads, and protect management interfaces and collected outputs.
Analysis useMalware sandboxes instrument a controlled environment to record behavior; they are one use of sandboxing, not the definition of every sandbox.
AssuranceTest escape paths, exposed interfaces, policy exceptions, resource limits, reset procedures, monitoring, and dependencies in the host or surrounding infrastructure.
Important limitationA sandbox reduces potential impact but does not prove code is safe. Vulnerabilities, unsafe exceptions, excessive permissions, shared resources, host compromise, or evasion can defeat isolation or conceal behavior, so defense in depth remains necessary.