AI Security and Privacy

AI Agent Sandbox Assurance

Examine the boundary between agent execution and the systems around it, including files, networks, secrets and cleanup.

Separate runtime capability from isolation evidence

Filesystem boundary
Map allowed mounts and protected host paths.
Network boundary
Identify permitted destinations and denied routes.
Secret boundary
Trace credential delivery and unintended exposure paths.
Lifecycle boundary
Examine resource limits, persistence and cleanup.
The scope, team and timing are agreed before an engagement begins.

Engagement boundary

Agree the work and its boundaries together

The engagement boundary

In scopeFilesystem boundary
Map filesystem, network, process and secret-access boundaries.
DependencyNetwork boundary
The inputs, authorised access and client participants are confirmed before scheduling.
Client responsibilityRuntime owner
Retains organisational decisions and risk acceptance.
BoundaryAdditional implementation
Changing the isolation platform, maintaining images and responding to newly disclosed vulnerabilities are separate work. Findings apply to the configuration and version examined, not every deployment of the product.
Only the agreed workflow, environments and activities form the engagement.

What the work covers

  • Map filesystem, network, process and secret-access boundaries.
  • Review isolation configuration and exercise authorised boundary tests.
  • Examine resource limits, persistence, cleanup and recovery behaviour.

Scope boundaries

  • Formal certification, legal conclusions and guarantees of safe or compliant behaviour.
  • Implementation or ongoing operation beyond the agreed scope.
  • Changing the isolation platform, maintaining images and responding to newly disclosed vulnerabilities are separate work. Findings apply to the configuration and version examined, not every deployment of the product.

Decision and fit

Separate runtime capability from isolation evidence

Compare the engagement with the decision you face

Useful decisionA question worth resolving
Separate runtime capability from isolation evidence.
NeededAn accountable participant
Infrastructure and security owners introducing code-execution or computer-use agents into controlled environments.
PrerequisiteInspectable source material
The relevant owners can make workflow evidence and agreed inputs available for examination.
Different needA scope mismatch
There is no agent execution environment or isolation boundary to review.
Discuss fit and prerequisites together; a different need may require a separately scoped engagement.

The engagement

Examine the boundary between agent execution and the systems around it, including files, networks, secrets and cleanup. An isolation review plus selected authorised boundary tests for a specified sandbox configuration.

The decision at stake

An agent can execute code, but the team lacks evidence about the isolation limits of its runtime.

Suitable situations

  • Infrastructure and security owners introducing code-execution or computer-use agents into controlled environments.

Less suitable situations

  • There is no agent execution environment or isolation boundary to review.

Working sequence

How the work moves from question to handover

What each stage should leave your team with

  1. Define the containment boundary

    Host, filesystem, network and secret boundaries.

  2. Review enforcement settings

    Runtime restrictions and resource limits.

  3. Exercise containment and cleanup

    Observed containment, exhaustion and cleanup behaviour.

  4. Define release checks

    Checks for future image or policy changes.

The sequence is tailored to the agreed workflow, evidence and acceptance criteria.

How the work proceeds

  1. Define the containment boundary. Map host access, mounted files, network egress, credentials and persistent state available to a workload.
  2. Review enforcement settings. Examine privileges, runtime restrictions, resource quotas and separation between concurrent tasks.
  3. Exercise containment and cleanup. Run agreed probes for forbidden access, exhaustion, persistence and residual data using disposable environments.
  4. Define release checks. Provide configuration findings and repeatable containment tests for image, policy or runtime changes.

Worked situation

Follow one situation through to a bounded next decision

Illustrative working situation

Illustrative scenario

  1. Starting condition

    A completed workload leaves data in a reusable workspace.

  2. Question to examine

    Inspect teardown and the next isolated execution.

  3. Possible next action

    Document the persistence finding and cleanup verification case.

Constructed example of the reasoning; no result or client engagement is claimed.

Decision artefacts

Leave a record the next owner can use

Isolation finding

Illustrative working format

Subject and version
Identify the workflow, configuration or proposal examined.
Filesystem boundary
Record the relevant requirement and supporting material.
Secret boundary
State the observation, supporting evidence and remaining uncertainty.
Runtime owner
Record the next decision, follow-up and unresolved question.
Illustrative structure; final deliverables are agreed for each engagement.

What the work can produce

  • Isolation boundary map
  • Sandbox control findings
  • Cleanup and containment verification cases

Working together

Your team and our team need a clear working agreement

Who contributes and who decides

Client sponsor
Confirms the decision, priorities and people needed to resolve it.
Client technical or process owner
Provides the agreed evidence, access and review of findings or deliverables.
Softlander delivery lead
Coordinates the agreed activities, review points, handover and open actions.
Authorised decision-maker
Accepts deliverables against the agreed criteria and separately decides operational or business action.
Client owners retain business, release and risk-acceptance decisions.

What the work depends on

  • Provide runtime and infrastructure owners, deployment configuration, network routes, secret-injection design and disposable test workloads. Agree resource limits, cleanup expectations and authorised probes before execution.

Engagement availability

The start date, named delivery roles, review points and fee basis are agreed after confirming the inputs and scope. Duration depends on evidence access, system complexity and the time your decision-makers need to review the work.

Engagement and handover

Agree the delivery model and what completion means

A clear agreement from scope to ownership

Scoped work
Map filesystem, network, process and secret-access boundaries.
Indicative handover
Isolation boundary map; Sandbox control findings
Review and acceptance
Review the agreed deliverables and record any open items, owners and conditions.
Ongoing work
Assign maintenance and operational ownership; agree additional scope and coverage explicitly.
The statement of work confirms milestones, review responsibilities, fees and change control before the engagement starts.

An isolation review plus selected authorised boundary tests for a specified sandbox configuration.

The handover maps intended isolation to configuration and observed test results, including what survives termination. Each finding names the affected host, tenant, network or resource boundary and a verification case.

Changing the isolation platform, maintaining images and responding to newly disclosed vulnerabilities are separate work. Findings apply to the configuration and version examined, not every deployment of the product.

Relevant reading

A little more clarity

Questions worth asking.

Start with the questions that shape a better decision.

When is AI Agent Sandbox Assurance useful?

An agent can execute code, but the team lacks evidence about the isolation limits of its runtime. Infrastructure and security owners introducing code-execution or computer-use agents into controlled environments.

How would we engage your team?

An isolation review plus selected authorised boundary tests for a specified sandbox configuration. The statement of work confirms the deliverables, review points, fee basis and responsibilities before work begins.

What does our team need to contribute?

Provide runtime and infrastructure owners, deployment configuration, network routes, secret-injection design and disposable test workloads. Agree resource limits, cleanup expectations and authorised probes before execution.

How do we accept the work and take ownership?

The handover maps intended isolation to configuration and observed test results, including what survives termination. Each finding names the affected host, tenant, network or resource boundary and a verification case.

What happens after the engagement?

Changing the isolation platform, maintaining images and responding to newly disclosed vulnerabilities are separate work. Findings apply to the configuration and version examined, not every deployment of the product.

How are timing and fees agreed?

We scope the work around the evidence and access available, technical complexity and the people needed for review. The proposal confirms milestones and whether fees are based on a defined scope, agreed delivery capacity or a recurring remit. Extra workflows, tests or integrations are agreed as changes rather than assumed to be included.

Discuss ai agent sandbox assurance

Tell us about the decision you are facing and we will confirm whether this service is the right starting point.