Skip to content
VITI Security

Cloud Pentesting: AWS, Azure, GCP — Lessons from Real Engagements

by VITI Security TeamMay 13, 2026
Cloud Pentesting: AWS, Azure, GCP - Lessons from Real Engagements - VITI Security

Cloud pentesting differs from network pentesting in scope and tooling. The "perimeter" is API endpoints and IAM. The blast radius is whatever the compromised role can do. Here is the methodology that consistently produces findings.

Engagement scoping

Cloud engagements need clear scope: which accounts, which regions, which assumed-breach starting point. The most common starting points are unauthenticated (find an open S3 bucket, find a public RDS instance), authenticated low-privilege user, or compromised CI/CD service principal.

Phase 1: Surface enumeration

  • Public buckets, blobs, and storage accounts (S3, Azure Blob, GCS).
  • Exposed serverless endpoints (Lambda function URLs, API Gateway with weak auth).
  • Public-facing services on standard ports — RDS, ElastiCache, public ALBs with weak rules.
  • Subdomain enumeration finds CloudFront distributions, S3 website endpoints, App Runner services.

Phase 2: Identity attack paths

Once a low-privilege identity is in hand, map what it can do. Tools: Pacu (AWS), MicroBurst (Azure), GCPBucketBrute (GCP). Look for over-permissioned policies, hidden role-chaining, and the classic "PassRole + iam:UpdateAssumeRolePolicy" privilege escalation paths. AWS IAM Access Analyzer is your friend.

Phase 3: Data and secrets

  • SSM Parameter Store and Secrets Manager — readable to over-broad roles in many environments.
  • S3 metadata — even when bucket contents are restricted, ACLs and policies often leak object existence.
  • EC2 instance metadata service (IMDSv1 still found) — credential theft via SSRF.
  • Cross-region forgotten resources — old Azure storage accounts, GCP projects from a 2020 PoC.

Phase 4: Lateral movement

  • Cross-account trust chains (assume-role into linked accounts).
  • Service-to-service via service identity (EC2 role → S3 → SSM → another EC2).
  • CI/CD as a pivot — many orgs grant CI roles broad permissions.

Findings that always appear

  • Public S3 buckets (or near-public — readable to "any AWS account").
  • Over-permissioned default IAM roles.
  • Long-lived access keys for human users.
  • Logging gaps: CloudTrail not enabled in all regions, or sent to a bucket that is itself attackable.
  • Forgotten test resources with elevated trust to production.

Reporting that lands

Cloud findings need clear "if this is exploited, here is what an attacker can do" narratives. CSPM tool output dumped to PDF is not a pentest report. A real cloud pentest documents the attack chain, the affected resources, and a prioritized remediation backlog with IaC examples where possible.

If you want a cloud pentest scoped for an SMB budget, our team runs them in 1-2 weeks across AWS, Azure, or GCP.

cloud pentesting

Further reading

Frequently asked questions

How is a cloud pentest scoped?

Cloud engagements need clear scope: which accounts, which regions, which assumed-breach starting point. The most common starting points are unauthenticated (find an open S3 bucket, find a public RDS instance), authenticated low-privilege user, or compromised CI/CD service principal.

How are identity attack paths tested in a cloud pentest?

Once a low-privilege identity is in hand, map what it can do. Tools: Pacu (AWS), MicroBurst (Azure), GCPBucketBrute (GCP). Look for over-permissioned policies, hidden role-chaining, and the classic "PassRole + iam:UpdateAssumeRolePolicy" privilege escalation paths.

What findings show up in almost every cloud pentest?

Public S3 buckets (or near-public - readable to "any AWS account"). Over-permissioned default IAM roles. Long-lived access keys for human users. Logging gaps: CloudTrail not enabled in all regions, or sent to a bucket that is itself attackable.