The cloud security best practices that prevent the most breaches are, in order: lock down identity, patch and inventory what faces the internet, close public data exposure, manage secrets, harden configuration, log and alert, and test the result.
Most cloud breaches are not sophisticated, and our cloud security statistics page has the measured causes. They come down to a handful of best practices that were skipped, misunderstood, or assumed to be someone else’s job. This is the practical list, in the order that actually reduces risk, drawn from what we find over and over in our cloud penetration testing services.
1. Get identity and access management right first
In the cloud, identity is the perimeter. Weak or absent credentials were the entry point in 27.2% of the cloud intrusions Google Cloud analyzed for the second half of 2025 (Threat Horizons Report, H1 2026), and over-privileged access decides how far an attacker gets.
- Enforce least privilege. Grant the minimum permissions a role needs, and nothing more. Wildcard permissions are the single most common finding we report.
- Eliminate unused credentials. Stale access keys and dormant accounts are free footholds for an attacker.
- Require MFA everywhere, especially for privileged and root accounts.
- Avoid long-lived keys. Use short-lived, scoped credentials and rotate what you must keep.
- Audit privilege escalation paths. A low-privilege foothold that can escalate to admin is the attack path we most often chain in testing.
If most of your cloud is Microsoft 365 or Google Workspace, the same identity rules apply to the tenant. A Microsoft 365 security assessment checks MFA coverage, admin roles and legacy sign-in there, and our Microsoft 365 security best practices guide lists the settings to change first.
For Google Workspace, the Google Workspace security checklist covers the same ground in the Admin console.
2. Patch and inventory internet-facing workloads
In the second half of 2025, vulnerabilities in third-party software were the entry point in 44.5% of the cloud intrusions Google Cloud analyzed, up from 2.9% in the first half and ahead of weak credentials (Threat Horizons Report, H1 2026).
- Inventory everything reachable from the internet, and the software running on it.
- Patch exposed software in days, not at the next maintenance window.
- Take down what you cannot patch, or put it behind a VPN or identity-aware proxy.
3. Protect your data
- Encrypt at rest and in transit with current, strong algorithms. Make it the default, not an opt-in.
- Close public exposure. No public storage buckets or blobs unless there is a deliberate, reviewed reason. Misconfigured storage leaks data with no exploit needed.
- Control backups and snapshots. They contain the same sensitive data as production and are frequently left exposed.
- Classify data so you know which systems deserve the most protection.
4. Manage secrets properly
- Get secrets out of code, environment variables, and metadata. Hardcoded keys are a recurring path to full-environment compromise.
- Use a secrets manager with rotation and scoped access. Build pipelines are where long-lived cloud keys most often leak, so see pipeline secrets and OIDC.
- Guard the metadata service. Server-side request forgery against cloud metadata endpoints is a classic escalation route, and so is a Kubernetes pod that can reach them; our Kubernetes security checklist covers blocking it. On AWS, require IMDSv2, the session-oriented metadata service, so a request without a token is refused (AWS documentation).
5. Harden configuration
- Change insecure defaults. Unnecessary services, default credentials, and permissive settings are low-hanging fruit.
- Restrict network exposure. Tight security groups and firewall rules; never leave management interfaces (SSH, RDP, databases) open to the internet.
- Benchmark against CIS foundations. CIS publishes a Foundations Benchmark each for Amazon Web Services, Microsoft Azure and Google Cloud Platform, and CSPM tools score against them. Use yours, but remember a benchmark checks settings while a test proves exploitability. A cloud security assessment runs that benchmark review by hand, read-only, from $4,200.
6. Monitor and detect
- Enable logging across the environment (control plane, data access, network) and make the logs tamper-resistant.
- Alert on the activity that precedes a breach: unusual privilege use, new access grants, and data access anomalies.
- Test that detection actually fires. Controls you have never exercised are assumptions, not defenses.
7. Test it, do not assume it
Every practice above is a control you believe is working. Testing is how you find out before an attacker does. Combine these layers:
- A cloud penetration test for the configuration and identity side, chaining misconfigurations into real attack paths rather than just listing them.
- Application and API testing for the code running in the cloud, which no configuration review will catch.
- Kubernetes penetration testing where workloads run on a cluster, to check what one compromised pod can reach.
We cover the reasoning behind testing the cloud account and the application together in more depth in our guide to cloud application security.
Cloud security checklist: every control in one table
| Control | AWS | Azure | Google Cloud | Verify | Priority |
|---|---|---|---|---|---|
| MFA for all users | Root user and IAM Identity Center | Conditional Access in Microsoft Entra ID | 2-Step Verification | Sign in without a second factor | 1 |
| Least privilege | IAM Access Analyzer | Privileged Identity Management | IAM Recommender | List who can administer the account | 1 |
| No long-lived keys | Roles, no old access keys | Managed identities | Service account keys disabled | Export keys, check ages | 1 |
| Exposed workloads patched | Patch Manager, Inspector | Update Manager | VM Manager | Scan your public addresses | 1 |
| No public storage | S3 Block Public Access | Blob anonymous access off | Public access prevention | Fetch an object anonymously | 1 |
| Metadata protected | IMDSv2 required | Narrow identity roles | Narrow service accounts | Test URL fetchers for SSRF | 2 |
| Logging on | AWS CloudTrail, all regions | Activity log export | Cloud Audit Logs | Find a test event | 2 |
Cloud penetration testing checklist: what a test covers
A configuration review ticks the table above. A cloud penetration test tries to break what the table assumes:
- IAM privilege escalation from a low-privilege identity to admin.
- Storage: public and cross-account access to buckets, snapshots and backups.
- Network exposure, from outside and from one compromised workload.
- Secrets in code, images, pipelines and the metadata service.
- Containers and serverless: escape to the node, cluster or account.
- Logging: whether the test itself was recorded and alerted on.
The code running in the account is a separate test, covered in our guide to cloud application security best practices.
The priority order
If you can only act on part of this list, do it in this sequence: lock down IAM, patch what faces the internet, close data exposure, manage secrets, then harden configuration and monitoring. That order maps directly to how real cloud attacks unfold.
Frequently asked questions
What is the first cloud security control a small company should fix? MFA on every human account, starting with the root or global administrator, then removing access keys nobody uses.
What does a cloud security assessment check? Configuration against the provider’s CIS benchmark, read-only: identity, storage, network rules and logging. A cloud penetration testing checklist goes further, as the section above shows.
How often should cloud configuration be reviewed? Continuously by tooling, and by hand at least once a year and after any major change.
Cloud gives you speed and scale, and an unfamiliar attack surface to go with it. If you run on AWS specifically, our guide to AWS penetration testing covers the IAM and storage attack paths in detail, the Azure penetration testing guide does the same for Entra ID, and the GCP penetration testing guide covers Google Cloud’s IAM and service accounts. If you want to know how your environment holds up against a real attacker rather than a benchmark, scope a cloud assessment and we will map the paths that actually matter. The cloud penetration testing cost starts at $6,800, fixed in writing.
Written by
Wahid Iqbal, Head of Compliance
20+ years in the field. Focus: PCI DSS, ISO 27001 & ISO 22301. Part of the team that scopes, tests, and reports for our clients. Full profile →