Cloud Infrastructure and Third-Party Authorization
A customer has authorized you to test an application running in the cloud. The application is clearly named in the statement of work, the test window is open, and you have valid credentials. You now want to place command-and-control infrastructure in the same cloud environment to support the assessment. Does the customer’s authorization cover that decision?
Not necessarily. The application test and the supporting infrastructure may involve different services, activities, and provider rules. The customer can authorize work on resources it controls, but it can’t grant permission over everything operated by the hosting provider. Your next decision therefore depends on more than the application’s target list. You need to know who controls each resource, which party can authorize the proposed activity, what the provider permits, and whether another reporting or approval requirement applies.
CompTIA PenTest+ exam PT0-003 places shared responsibility and legal and ethical considerations within pre-engagement activities. For this lesson, the important point isn’t simply remembering that those subjects appear in the objective. It’s understanding why they affect a technical choice before the test begins.
The shared responsibility model divides security responsibilities among the cloud provider, the customer, the tester, and sometimes other parties. It doesn’t mean every responsibility is split evenly. It also doesn’t produce one universal dividing line for every cloud service.
The boundary can vary according to the service being used. A customer may manage an application, its data, identities, or configuration while the provider operates underlying infrastructure. With a different service, the division may be different.
The tester has a distinct role in that arrangement. You’re responsible for staying within the authority granted to you, following the engagement constraints, and choosing validation methods whose impact is justified. If the customer uses a consultant, managed service provider, software vendor, or another third party, that party’s access and authority must also be established. The label third party doesn’t answer whether testing is allowed. Written evidence does.
An authorization letter is written evidence that a tester is permitted to conduct defined security testing. A useful authorization identifies enough of the work to connect the person or testing organization to the targets, activities, dates, and conditions that were approved. It shouldn’t be treated as a magical permission slip. A letter from the customer can establish the customer’s approval, but it doesn’t silently override a cloud provider’s rules, expand the customer’s ownership, or authorize testing against a neighboring tenant.
This is where provider-specific policy becomes part of pre-engagement planning. Amazon Web Services permits many customer security assessments without a request for prior approval, but that doesn’t mean anything hosted on AWS is fair game. Testing AWS infrastructure itself is restricted, and command-and-control related testing requires prior approval. Those distinctions separate testing an authorized customer resource from conducting an activity governed by an additional provider requirement.
So, evidence that a customer owns an account or pays for a subscription isn’t enough. Evidence that an application is in scope also doesn’t prove that every supporting technique is allowed. You need the current provider policy for the actual service and activity, the customer’s authorization, and the engagement documents that define your targets and constraints. A general memory that cloud testing is allowed is weaker than those specific records.
A reliable decision process starts with the requirement. What outcome has the customer authorized you to evaluate? Next, predict what should be true if the proposed action is permitted.
You’d expect the named target to fall within the customer’s authority, the activity to fit the engagement, and any provider-specific condition to be satisfied. Then choose a bounded action to confirm those expectations. Finally, compare the evidence with your prediction and document both the decision and any uncertainty.
Apply that process to the consultant who wants to host command-and-control infrastructure. The authorized outcome is an assessment of the customer’s application. The first target is therefore the application and the resources explicitly included with it. Hosting command-and-control infrastructure is a separate proposed activity. Because AWS identifies command-and-control related testing as requiring prior approval, the consultant shouldn’t infer permission from the application authorization alone.
What evidence should exist before proceeding? At a minimum, the consultant needs clear customer authorization covering the relevant work, a target inventory that shows which resources belong to the engagement, timing and operational constraints, and evidence that the provider’s additional approval requirement has been satisfied. If the documents don’t establish those points, the defensible next action is clarification or escalation. It isn’t silently broadening the meaning of the original authorization.
Notice that this decision doesn’t require launching anything to discover whether it’s allowed. Reviewing the authorization, target inventory, provider policy, and approval status is lower impact and directly relevant. Penetration testing often rewards technical initiative, but initiative doesn’t mean acting first and reconstructing permission afterward. When documentary evidence can answer the authorization question, that’s the appropriate first validation.
The same discipline applies to Microsoft Azure, but you shouldn’t assume the wording or conditions are identical. Azure permits authorized testing of resources you own without pre-approval. When a third party conducts the test, written authorization is required, and Azure’s rules of engagement control the activity. A customer saying, “Our consultant is approved,” may express intent, but the required written authorization is the stronger evidence.
Before choosing a technique in either environment, identify five things. Determine who authorized the test, which target the technique will touch, when it may run, what other systems or services it depends on, and what operational effect it could have. These aren’t administrative details detached from the technical work. They determine whether the technique is a valid action at all.
Suppose a technique is technically capable of validating a weakness, but it could affect resources outside the named application. A less disruptive method could establish the same evidence without crossing that boundary. The lower-impact method should come first. A stronger technique becomes appropriate only when it remains authorized, its added value is clear, and the relevant operational constraints allow it. Technical effectiveness outside the rules of engagement is not technical correctness for the engagement.
Mandatory-reporting requirements belong in this decision process as well. Mandatory doesn’t mean that every cloud test has one universal reporting rule. It means an applicable requirement makes reporting obligatory rather than optional.
The source of that requirement might be found in the engagement documents, governing rules, or another applicable obligation. The tester must identify what condition triggers reporting, who must receive it, and what approved process applies. Those specifics can’t safely be guessed from the phrase mandatory reporting alone.
Imagine that a proposed test could reveal exposure involving a third-party service. Before conducting it, you’d check whether the engagement defines an immediate notification or escalation requirement. If it does, finding collection and communication need to support that requirement. If the documents are unclear, that ambiguity is itself a pre-engagement issue to resolve. You shouldn’t invent a recipient, deadline, or legal conclusion, and you shouldn’t assume the final report is always the first required communication.
Tester risk connects all these questions. Risk to the tester can include acting without adequate authority, affecting systems outside scope, disrupting operations, or handling evidence in a way that conflicts with an engagement obligation. The proper response isn’t to abandon useful testing. It’s to reduce uncertainty before acting, preserve proof of authorization, and select a method proportionate to the evidence needed.
Now consider incomplete evidence. A consultant has written customer authorization for an Azure application. The authorization names the application but not the consultant’s testing company, and the target list includes an address whose ownership is uncertain. The consultant also has a tool result that appears to identify a serious weakness. There are several problems, and the tool result doesn’t erase any of them.
First, Azure requires written authorization for a third party, so the consultant should verify that the writing actually establishes the third party’s authority. Second, the uncertain address must be tied to an owned and authorized resource before stronger testing touches it. Third, the observed result needs validation. A result associated with the wrong target, stale data, insufficient privileges, a missing prerequisite, or a configuration mistake may not support the conclusion the tester first drew.
The next step is not automatically a more aggressive test. Start by checking the target identifier, ownership information, relevant configuration, testing prerequisites, account privileges, and the freshness and quality of the available data. Ask what result the tool should have produced under the stated conditions.
Then compare that prediction with the observation. If they conflict, preserve the result as an observation while you investigate. Don’t promote it to a validated finding merely because the output looks severe.
This distinction matters in cloud environments because an address, name, or service endpoint doesn’t prove ownership by itself. Cloud resources can involve provider-managed components, customer-managed components, and dependencies owned by others. An observation may be genuine while your assumption about the responsible or authorized party is wrong. The authorization evidence and target inventory must agree with the technical evidence.
A common and risky assumption is that owning a cloud subscription means every provider service, tenant, or neighboring resource can be tested. It doesn’t. A subscription establishes a customer relationship and control over certain resources. It doesn’t transfer authority over provider infrastructure or another customer’s environment. Nor does it convert every activity into an approved one.
How would you expose that mistaken assumption before causing harm? Ask for evidence connecting the proposed target to the customer’s owned resources and to the engagement’s explicit scope. Then check the provider’s current rules for that service and activity.
If you can prove customer ownership but not engagement scope, you still need scope clarification. If the engagement includes the resource but the provider restricts the activity, customer approval alone is still insufficient. Each layer answers a different question.
A failure to get a response from a target doesn’t prove that the provider blocked the test, and an unexpected response doesn’t prove that you found a vulnerability. Either result may reflect an incorrect target, missing permission, account privilege, configuration, prerequisites, or data quality. Troubleshooting should revisit those assumptions before escalating to a stronger technique. That protects the environment and improves the credibility of the eventual report.
When a low-impact verification confirms the target and the expected behavior, record what it establishes and what it doesn’t. You might be able to say that the authorized application produced a particular response under defined conditions. You might not yet be able to say that an underlying provider service is vulnerable. Careful reporting keeps the conclusion aligned with the evidence and with the responsibility boundary.
The practical product of this work is a defensible decision record. It connects the customer’s goal to the party granting authorization, the exact target, the provider’s applicable rules, the selected technique, the anticipated impact, and any reporting condition. That record supports planning before testing, analysis during testing, and accurate communication afterward. It also makes unresolved questions visible instead of burying them in an assumption.
Consider two plausible choices in another hypothetical engagement. A customer authorizes testing of its public application on AWS. The tester discovers that a validation technique might send activity toward a provider-operated component not listed as a customer target. One choice is to run the technique because the customer owns the cloud account. The other is to pause, verify the component’s ownership and scope, review the provider rule that applies to the activity, and select a lower-impact validation if one can establish the needed evidence.
The second choice is stronger. The first rests on the false idea that account ownership creates unlimited testing authority. The second separates customer authorization from provider restrictions and asks whether the proposed action is both useful and permitted. It also creates an opportunity to identify any reporting or escalation requirement before the action generates sensitive evidence or operational impact.
Would the answer change if the component were conclusively shown to be a customer-owned resource and explicitly listed in scope? That would resolve important target and customer-authorization questions, but you’d still check the provider’s rules for the specific activity. If those rules permit the work and the timing and operational constraints are satisfied, a bounded test may be justified. If the activity requires additional approval, the fact that the target is in scope doesn’t remove that requirement.
Now change the platform to Azure and make the tester an outside consultancy. The consultancy has an email from an employee saying the work is fine, but the rules require written authorization for the third party. The decision turns on whether the available writing is valid authorization from an appropriate customer authority and whether it adequately covers the consultancy, resources, activities, and timing. An informal message that lacks those elements shouldn’t be stretched to cover them. The tester needs the authorization gap corrected before proceeding.
These comparisons show why memorizing “cloud penetration testing doesn’t need pre-approval” is unreliable. Both AWS and Azure permit categories of authorized customer testing without routine pre-approval, yet each places conditions on what can be tested and how. AWS calls out restrictions on testing its infrastructure and prior approval for command-and-control related testing. Azure emphasizes owned resources, written authorization for third-party testers, and compliance with its rules of engagement. You must check the policy for the platform, service, party, and activity actually involved.
During the engagement, keep testing decisions tied to the same evidence. New information can change the safe and authorized next step. A newly identified dependency may belong to another party. A planned technique may have a greater operational effect than expected.
A finding may trigger a required notification process. When that happens, don’t treat the original authorization as infinitely elastic. Recheck the boundary, document the new condition, and seek clarification or escalation where necessary.
This habit also improves reporting quality. A report should distinguish what was authorized and tested from what was merely observed or left unresolved. If a provider restriction prevented a particular validation, say that the condition wasn’t validated within the authorized method rather than claiming the underlying weakness was confirmed. If target ownership couldn’t be established, preserve the relevant observation and explain the limitation. Clear limitations are more useful than certainty the evidence can’t support.
Cloud shared responsibility therefore isn’t only a diagram showing provider and customer duties. In a penetration test, it becomes a chain of authority and evidence. The customer defines and authorizes work within its control.
The provider’s current rules constrain activity on its platform. The tester applies both sets of boundaries, manages operational impact, validates results, and follows applicable reporting requirements. Third parties need authority appropriate to their role rather than borrowed assumptions.
Before the first technique runs, you should be able to explain whose resource is involved, who has authority to approve the work, which provider condition applies, what evidence you expect, and how you’ll validate it with the least necessary impact. You should also know what uncertainty would make you pause and what reporting or escalation path applies if the test reveals a triggering condition. Those answers turn a technically possible action into a professionally defensible one.
A cloud testing decision is defensible only when its layers agree. The customer must have authority over the named resource, the tester must have written permission appropriate to the work, and the provider’s rules must permit the specific activity. A tool result can’t replace any of that evidence.
Validate ownership, scope, expected behavior, and operational impact before choosing a stronger technique. If an authorization boundary or reporting requirement is unclear, clarification is the correct next action, not an assumption. Following that pattern lets you connect shared responsibility to real testing choices while protecting the customer, the provider, third parties, and yourself.
Test your recall with this decision: Does the customer’s authorization cover that decision? The matching episode reference preserves exact terminology, named tools or standards, source-derived evidence checkpoints, and review material for Cloud Provider Authorization and Shared Responsibility. This lesson was produced by Bare Metal Cyber dot com.