← Back to blog

Google Cloud penetration testing: scope, rules and cost

October 23, 2026
Google Cloud penetration testing: scope, rules and cost

A Google Cloud penetration test is a controlled, authorised attack on your own Google Cloud setup: your projects, identities, storage, networks and the applications you run there. Google secures the underlying platform. You secure what you build on it, and that is the part a tester examines. If you are a UK startup, SME, school or charity, this guide explains what is in scope, what is not, how to prepare, and how to choose a tester without paying for more than you need.

Solusec
solusec.co.uk
Find vulnerabilities in your cloud setup
Solusec provides expert-led penetration testing with clear, actionable findings to help you strengthen the security of your Google Cloud environment.
Get a penetration test quote

Key takeaways

  • Google looks after its own infrastructure. Your configuration, user access and application code remain your responsibility, and that is what a test targets (NCSC penetration testing guidance).
  • Scoping decides the value of the test. Agree boundaries, sensitive systems and test scenarios with your risk owners, technical staff and the lead tester before any testing starts.
  • Choose a CREST-accredited firm and ask who will actually do the work. Automated scans alone miss chained, logic-based attacks.
  • Treat testing as a cycle. Cloud environments change quickly, so a test at one point in time ages fast.

Table of Contents

What a Google Cloud penetration test actually covers

A cloud test is not a test of Google. It is a test of your use of Google. A competent tester starts from the same position as an attacker who has found a foothold, whether that is a leaked key, a phished employee, a public storage bucket or a vulnerable web application, and tries to see how far it goes.

In practice the work usually falls into these areas:

  • Identity and access management (IAM). Over-broad roles, service accounts with more rights than they need, long-lived keys, and ways a low-privilege user could climb to a higher one. Google Cloud organises resources in a hierarchy of organisation, folders, projects and resources, so a permission granted high up can quietly reach a long way down.
  • Storage and data exposure. Cloud Storage buckets, database snapshots and backups that are more open than intended.
  • Network exposure. Firewall rules, public IP addresses, load balancers and services reachable from the internet that should not be.
  • Compute and containers. Virtual machines, Kubernetes clusters and serverless functions, including how secrets and credentials are handled inside them.
  • Your applications. Web apps and APIs hosted on the platform, tested for the usual flaws such as broken access control and injection, plus how they interact with cloud services.
  • Logging and detection. Whether suspicious activity would actually be noticed.

The same breakdown answers a common search, "Google cloud security testing": it can mean a configuration review, a penetration test or both. They are different jobs. A review checks settings against good practice. A penetration test tries to exploit what it finds and shows the real consequence. If you are unsure which you need, our short explainer on penetration testing versus vulnerability scanning sets out the difference.

Who is responsible for what

The most expensive misunderstanding in cloud security is assuming the provider has covered it. The NCSC is clear that with infrastructure-as-a-service you remain responsible for secure configuration, user access control and firewall management, and that the provider's platform security does not replace your own checks (NCSC).

The split looks like this:

AreaGoogleYou
Data centres, hardware, hypervisorYesNo
Core platform servicesYesNo
IAM roles and who holds themNoYes
Storage permissions and encryption choicesTools providedYes
Firewall and network configurationTools providedYes
Your application code and dependenciesNoYes
Monitoring, alerting and responseTools providedYes

This is why a good scope excludes Google's own infrastructure and includes your misconfigurations. Get this wrong and you either waste money testing things you cannot change, or leave the genuine risks untouched.

Do you need Google's permission?

This is the question behind many "GCP penetration testing" searches. The honest answer is to read the current rules rather than rely on a blog's summary, because they can change and some managed services have their own procedures. Before any test, read Google Cloud's Acceptable Use Policy and the terms covering the specific services in scope, and have your tester confirm in writing that the planned activity is permitted.

The NCSC also advises checking your legal and technical contracts for explicit consent requirements before a third party probes your environment, because some automated tools can breach provider terms or trigger defences that lock accounts (NCSC). Two practical steps follow:

  1. Make sure the person signing the authorisation actually owns the projects being tested. If a supplier or an MSP hosts part of your environment, get their consent too.
  2. Name the exact projects, domains and IP ranges in a signed scope document, so nobody is in doubt about what is authorised.

How to scope the test properly

The NCSC treats scoping as the step that matters most. Its guidance is that the scoping discussion should include the organisation's risk owners, technical staff and the lead tester, with clear boundaries, sensitive systems identified and test scenarios agreed (NCSC). A sensible scoping conversation answers these questions:

  • What are you protecting? Customer data, a payments flow, student records, donor information. Test the route to those assets first.
  • Which projects and environments are in? Production, staging, development. Many breaches begin in a forgotten development project.
  • What is the attacker's starting point? An outsider with no access, an authenticated user, or a compromised developer laptop with a service account key. Each tells you something different.
  • What is off limits? Fragile production databases, third-party services you do not control, and anything covered by separate provider rules.
  • When can testing happen? If a test might disrupt services, arrange out-of-hours windows.
  • What will you whitelist? Give your tester's source IP addresses to your security tooling so the test measures your real defences instead of being blocked by your own firewall (NCSC).

Use Google's own tools first

Google Cloud ships tools that find many misconfigurations cheaply. Security Command Center surfaces misconfigurations and threats, and Security Validation in Google Security Operations lets you simulate attacks in your own Google Cloud environment. Fix what these report before the external test. Then the paid testers spend their time on the deeper, chained problems that automation cannot reason about, rather than on findings you could have closed yourself.

Why automated scanning is not enough

The NCSC's position is that an external team gives you an objective, expert view of your vulnerability status, and that organisations should not rely only on internal automated scans, which tend to miss complex chained exploits a human tester finds by imitating real attackers (NCSC).

A typical cloud chain looks like this. A low-risk web flaw lets an attacker read a file on a virtual machine. That file holds a token for a service account. The service account has an editor role on the project. The project holds a storage bucket with customer exports. No scanner reports "customer data exposed", because each step looks minor alone. A tester puts the steps together and shows you which single change breaks the chain.

Exploit chain from web flaw to exposed data

How often to test

Government guidance for digital services is to run penetration tests at least every 12 months, and to discuss significant changes with the security team before and not after (The GDS Way). That is a sensible baseline for a private company too. Add a test when something material changes: a new product launch, a migration, a major re-architecture, or a move to a new Google Cloud organisation structure.

If you deploy infrastructure as code and release several times a week, a yearly snapshot leaves long gaps. In that case consider a rhythm of lighter, more frequent testing alongside the annual deep test. Our pages on annual penetration testing and continuous penetration testing describe the options.

What a good report should give you

A test is only useful if you can act on the result. Look for:

  • A plain-English summary a director or trustee can read in five minutes.
  • Findings ranked by real risk to your business, not just a generic severity score.
  • Evidence for every finding: what was done, what was reached, and how to reproduce it.
  • Specific fixes, ideally with the exact IAM role, firewall rule or code change.
  • A chance to talk to the person who did the testing.
  • A retest to confirm your fixes worked.

If a quote does not mention these, ask. The most common complaint about cheap tests is a scanner export with a cover page.

Choosing a CREST-accredited tester in the UK

The NCSC recommends commissioning penetration testing from a suitably qualified, CREST-accredited firm for cloud environments (NCSC). Accreditation sets a floor. Beyond it, ask these questions:

  1. Who will test my environment? Named, experienced people, not a junior running a tool.
  2. Have they tested Google Cloud specifically? Google's IAM hierarchy, service accounts and metadata service behave differently from other clouds, so experience on one provider does not transfer automatically.
  3. How do they handle authorisation and data? Signed scope, secure evidence handling, and clear rules on production data.
  4. Do you get a retest? Ask whether it is included.
  5. Can you speak to them before buying? Our own approach is that you deal with the testers and assessors directly, not a sales team.

What it costs

Cloud tests are scoped by effort, so there is no single price. As a reference point, our published fees for two common fixed-scope tests are £3,000 for an external IP test and £2,500 for a CMS website test, both one-off. A Google Cloud test covering IAM, storage, compute and applications is scoped to your environment, so we agree the work and fee with you after a scoping conversation. The main cost drivers are the number of projects, how many applications and APIs are in scope, whether the test is authenticated, and whether you need a retest.

How Solusec can help

We are a CREST-accredited penetration testing company working with startups, small businesses, schools, colleges, universities and charities across the UK. You speak directly to the people doing the testing, and the report tells you what to fix first in plain English. See our penetration testing services for the fixed-scope options, and our cloud security assessment page for how we approach cloud environments. Tell us you run on Google Cloud and we will scope it with you. If you would rather have testing on a rolling basis, look at our managed penetration testing.

If you also need to show customers or funders a baseline, we are a certification body for Cyber Essentials. The scheme's requirements say that cloud services holding your data or services must be in scope (NCSC requirements v3.3), so your Google Cloud tenant matters there too. Our Cyber Essentials page lists certification from £320 to £600 depending on organisation size, and the NCSC itself says prices for standard certification start from £320 plus VAT.

Preparing for the test: a checklist

  • Export a list of projects, folders and who owns each.
  • Remove unused service account keys and users who have left.
  • Run Security Command Center and fix the high-severity findings.
  • Agree the testing window and an emergency contact on both sides.
  • Whitelist the tester's IP addresses in your security tooling.
  • Take backups of anything fragile in scope.
  • Tell your internal team and any managed service provider, so a test is not mistaken for a real attack.

Frequently asked questions

Is Google Cloud penetration testing allowed?

Generally you can test your own resources, provided you follow Google's Acceptable Use Policy and the terms for the services involved. Some managed services have their own procedures, so check the current rules and have your tester confirm the plan in writing before starting.

Is a vulnerability scan the same as a penetration test?

No. A scan lists known weaknesses automatically. A penetration test has a person try to exploit them and chain them together, which the NCSC says scans often miss (NCSC).

How long does a Google Cloud penetration test take?

It depends on the number of projects, applications and APIs in scope. Small environments are often a matter of days of testing plus reporting time. We confirm the timetable at scoping so you can plan around releases.

Does my small business really need one?

The NCSC reports that around one in two UK small businesses suffers a cyber incident each year, and that simple measures reduce the risk, while more complex cloud deployments warrant independent expert testing (NCSC). If you hold customer, student or donor data in Google Cloud, it is worth doing.

Sources

This article was created with assistance of AI. We never use AI with your data unless given explicit permission.