Penetration testing scope that makes authorization actionable
Penetration testing scope defines exactly which assets, identities, environments, data, actions, and time windows are authorized. A useful scope is operational: an engineer or AI agent can use it to decide what may be tested, what must stop, who approves a change, and what evidence is required.
Record the asset inventory, starting identities, environments, exclusions, sensitive-data rules, permitted and prohibited actions, testing windows, escalation contacts, stop conditions, evidence handling, and change approvals. Attach those terms to the statement of work and the execution system.

What is penetration testing scope?
Penetration testing scope is the written boundary of an authorized assessment. It identifies the systems and origins to be tested and the actions that are allowed. It also states exclusions, environmental limits, data rules, communication paths, and the conditions that stop the engagement.
Scope is not only a list of domains. A complete scope distinguishes production from staging, external from internal, user roles from administrative roles, read-only actions from state changes, and test data from real customer records.
The scope must be usable during execution. If a tester or autonomous agent has to ask which of two similar hosts is authorized, the scope is incomplete. If an action can affect customers but the document does not define approval, the scope is unsafe.
Both the customer and the testing provider should approve the final scope. The statement of work, authorization letter, rules of engagement, and technical execution configuration should describe the same boundary.
What should a penetration testing scope document include?
Include the engagement objective, business owner, testing provider, authorization signer, asset inventory, environments, test identities, permitted actions, exclusions, data handling, schedule, communication plan, stop conditions, evidence retention, and retest expectations.
Add the technical method used to identify an asset. A domain may resolve to multiple IPs, a cloud account may contain shared services, and an API may call third-party systems the customer does not own. Scope should state whether discovered dependencies are excluded or require approval.
Define version control for scope changes. Record the approved version in the testing system and attach each new asset or action to an amendment. An email thread is difficult to enforce when an automated agent can make decisions at machine speed.
| Field | Required content | Execution control |
|---|---|---|
| Objective | Business decision, risk focus, and assessment type | Test selection and reporting audience |
| Assets | Exact domains, IPs, applications, APIs, accounts, and environments | Target allow list and deny list |
| Identities | Roles, tenants, credentials, MFA state, and token scope | Authenticated test boundaries |
| Actions | Read, write, destructive, credential, and social actions | Tool policy and approval thresholds |
| Data | Synthetic data, sensitive records, extraction, and retention | Redaction and storage rules |
| Schedule | Windows, blackout periods, contacts, and escalation | Rate limits and stop conditions |
How should teams define assets and environments?
List assets by ownership and type: web applications, APIs, mobile clients, network ranges, cloud accounts, Kubernetes clusters, identity systems, data stores, and third-party integrations. Use immutable identifiers where possible, not only product names.
Separate production, staging, development, and lab environments. A production asset may be in scope for read-only validation while a staging replica is approved for destructive exploitation. Scope should state that difference explicitly rather than applying one action policy everywhere.
Resolve shared infrastructure. Load balancers, CDNs, cloud metadata services, managed databases, email providers, and authentication vendors may support the target but not belong to the customer. Mark shared components, allowed interactions, and any provider-specific testing rules.
Attach each asset to an owner and an approval path. If an in-scope application depends on an out-of-scope service, the test plan should state which requests can reach it and how to avoid affecting unrelated customers.
Which identities and data should be authorized?
Provide dedicated test identities for the minimum roles and tenants needed to evaluate the engagement objective. Document permission sets, ownership relationships, synthetic records, MFA expectations, token lifetime, and the person who can rotate or revoke each credential.
Do not authorize real customer data by default. If production records must be used, define the legal basis, minimum necessary access, masking rules, storage location, retention, and deletion. The agent should prove access with the smallest possible data sample.
Define how credentials are transferred and revoked. Avoid sending secrets through chat, tickets, or shell history. Use a scoped secret store, record which test identity used each credential, and revoke access after closure.
Identity scope should include service accounts and workload identities, not only human users. Broad cloud roles, CI tokens, MCP servers, and external tool credentials can create a larger blast radius than an ordinary user account.
- Dedicated users, tenants, service accounts, and synthetic records.
- Minimum permissions required for each planned test.
- Explicit MFA, token, and session handling rules.
- Credential rotation and revocation owners.
- Sensitive-data limits and evidence-redaction requirements.
What should rules of engagement prohibit or require?
Rules of engagement translate scope into action. Define allowed rates, prohibited techniques, destructive-action thresholds, social-engineering limits, data-extraction caps, third-party restrictions, and the exact approval needed for a state-changing test.
Name the people who can approve an exception and how quickly. Include a technical contact, incident-response contact, legal or privacy contact, and an executive stop-contact. The provider and customer should test the contact path before the engagement.
Define stop conditions: unexpected service degradation, exposure of real data, unapproved lateral movement, authentication lockouts, out-of-scope discovery, or a tool behaving outside policy. The run should pause, preserve evidence, and wait for a decision.
For autonomous or AI-driven testing, encode these rules outside the model. Asset allow lists, network policy, tool permissions, output limits, and deterministic approvals matter more than a prompt that says “stay in scope.”
How should scope connect to the statement of work?
The statement of work should reference the approved scope version and attach the asset inventory, test matrix, exclusions, rules of engagement, deliverables, schedule, and change procedure. Avoid duplicating conflicting details across separate documents.
Define what counts as in scope and who decides when discovery changes it. Record the amendment process, approvers, effective date, asset additions, removed targets, changed actions, and any fee or schedule impact.
Keep the authorization evidence with the execution record. A task should show which scope version allowed each asset and action. When a finding enters remediation, its evidence should reference the same authorization context.
- 01Draft
Record objective, assets, identities, environments, data, actions, and exclusions.
- 02Review
Have security, engineering, provider, privacy, and legal owners review the boundary.
- 03Approve
Capture named authorization and attach the exact scope version to the SOW.
- 04Enforce
Apply target, network, credential, rate, and action controls in the testing tool.
- 05Amend
Require explicit approval before adding assets, actions, data, or environments.
- 06Archive
Retain scope, approvals, changes, and evidence with the engagement record.
What are the most common penetration testing scope mistakes?
The most common failure is a list of domains without identities, environments, or action rules. The second is assuming that “the application” includes every API, cloud service, third party, and internal host it calls. The third is allowing state-changing tests without rollback or approval.
Another failure is stale scope. Domains move, cloud accounts change, new tenants appear, and credentials rotate. If the execution system was not updated, a previously authorized asset may now belong elsewhere, or a test identity may have gained excessive permissions.
Treat scope as a living control. Revalidate the asset ownership and authorization before a recurring run, and require a new approval when the business objective, data boundary, or destructive-action policy changes.
- Are exact assets and environments enumerated?
- Are test identities and permissions documented?
- Are destructive and state-changing actions classified?
- Are data access, extraction, and retention rules explicit?
- Are third-party and shared services excluded or approved?
- Are stop contacts and escalation paths tested?
- Is the executing system configured to the approved scope version?
Frequently asked questions
What is penetration testing scope?
It is the written boundary of authorized testing: assets, identities, environments, actions, data, time windows, exclusions, approvals, stop conditions, and evidence rules.
What should a penetration testing scope of work include?
Include the objective, asset inventory, test identities, environments, permitted and prohibited actions, sensitive-data handling, schedule, contacts, escalation, deliverables, evidence retention, and change procedure.
Can production be included in a penetration test?
Yes, with explicit authorization, low-impact test rules, rate limits, isolated data, rollback plans, and approval for state-changing or impact-bearing actions.
How should new assets discovered during testing be handled?
Treat them as out of scope until the customer approves an amendment that identifies the asset owner, test authorization, actions, data rules, and schedule impact.
How should AI agents enforce penetration testing scope?
Enforce target allow lists, network restrictions, credential scopes, tool permissions, action approvals, rate limits, and stop conditions outside the model, and record the approved scope version in every task.
References and review notes
Start with source. Test with authorization.
Review the public ARTEX repository, run it in an owned lab, and keep human approval on actions that can change a target.
ARTEX