AWS Cloud Pentesting
Purpose
Full AWS offensive methodology: from an identified AWS account/principal to mapped permissions, discovered resources, exposed data, and escalation/impact paths. Works black-box (via discovery chains) and credentialed (operator-supplied or discovered credentials).
Entry Conditions
- Black-box: account IDs/bucket names/regions from
cloud-attack-surface-discoveryor an SSRF→metadata chain - Credentialed: validated credentials from
cloud-credential-abuse(aws sts get-caller-identityalready recorded)
Discovery Signals
- 12-digit account IDs (presigned URLs, S3 ARNs, Lambda invocation error leaks)
- Bucket names, API Gateway IDs, Cognito pool IDs in client-side code
- Region hints from endpoint hostnames
Attack Model
identity/context established
→ permission enumeration (watch AccessDenied GAPS — they hide paths)
→ resource discovery across ALL regions
→ data exposure + credential harvesting
→ escalation candidates
→ verified impact
Methodology
Step 1: Identity and Boundaries
aws sts get-caller-identity
# user vs assumed-role; root keys: enumerate only, never destructive actions
aws iam list-attached-user-policies --user-name <u>
aws iam list-user-policies --user-name <u> # inline
aws iam get-account-summary # global account posture hints
aws iam simulate-principal-policy --policy-source-arn <arn> --action-names "*"
Step 2: All-Region Resource Discovery
IAM, S3, and Secrets Manager naming are global; most everything else is regional. Region-scoped enumeration silently misses us-west-2/eu-* resources:
for r in us-east-1 us-east-2 us-west-1 us-west-2 eu-west-1 eu-central-1 ap-southeast-1 ap-southeast-2 ap-northeast-1 sa-east-1; do
export AWS_DEFAULT_REGION=$r
aws ec2 describe-instances --query "Reservations[].Instances[].InstanceId" --output text 2>/dev/null
aws lambda list-functions --query "Functions[].FunctionName" --output text 2>/dev/null
aws ecs list-clusters --output text 2>/dev/null
done
aws s3api list-buckets # global
aws secretsmanager list-secrets # global
aws ssm describe-parameters # global
aws ecr describe-repositories --region us-east-1
Step 3: Deep Enumeration on Findings
- Lambda:
get-function-configuration(env vars, handler, role), download the deployment package (aws lambda get-function→ Code Location URL) when scope permits — packages routinely contain.envfiles and hardcoded keys - ECS: task definitions carry env vars/secrets and task-role ARNs
- EC2:
describe-instances(instance profiles, user-data), security groups, snapshots - Secrets Manager/SSM: names →
cloud-secrets-data-access - KMS: key policies (
get-key-policy) as trust surfaces →cloud-cross-account-tenant-trust - Cognito: user pool configurations, app clients, identity pools (client IDs are in frontend JS)
- CloudFormation:
describe-stacksreveals parameters (often secrets in plainParameterValueif NoEcho was forgotten) - Snapshots/AMIs: public or shared-with-account exposures (
describe-snapshots --restorable-by-user-ids allfrom credentialed context)
Step 4: Resource-Based Policies
S3/KMS/SNS/SQS/ECR/Lambda resource policies grant access identity policies never show:
aws s3api get-bucket-policy --bucket <b>
aws ecr get-repository-policy --repository-name <r>
aws kms get-key-policy --key-id <k> --policy-name default
Step 5: Chain
- Escalation candidates →
aws-iam-privilege-escalation - Storage exposure →
cloud-storage-exposure-testing - Secrets →
cloud-secrets-data-access - Trust edges →
cloud-cross-account-tenant-trust - EKS boundary →
container-securityfor internals, IRSA/node roles here
Tool Policy
AWS CLI + Python are the baseline. Optional accelerators:
- Pacu:
iam__enum_permissions(complete first — privesc_scan reasons only over collected permissions; watch for AccessDenied oniam:GetAccountAuthorizationDetailswhich silently degrades results),set_regions all, theniam__privesc_scan,lambda__enum(env vars),ssm__download_parameters,secretsmanager__enum,ebs__enum_snapshots_unauth - CloudFox:
aws permissions,env-vars,whoami— fast blast-radius and permission listing - ScoutSuite: credentialed posture sweep — grep the run log for AccessDenied before trusting "clean" results
- All tool output is HYPOTHESIS, not proof — see Evidence Contract
Evidence Contract
candidate resource/permission (tool output or list call)
→ effective-access check (actual GET/read or simulate-principal-policy)
→ concrete data retrieved / action performed
→ report
NOT evidence: Pacu/ScoutSuite/Prowler findings without live reproduction, a 200 from a create-style call with no read-back, silent clean runs (permission gaps), a list-only view of resources with no access demonstrated
Common Misses
- AccessDenied gaps during enumeration treated as "no resources" instead of "uncollected"
- Single-region enumeration
- Deployment packages/layers never downloaded for embedded secrets
- Resource-based policies never read (identity-side-only view)
- CloudFormation parameters and ECS task definitions skipped as secret sources
- Cognito misconfigurations (insecure app-client flows, identity-pool role mappings)
False Positives / Non-Findings
- Resources in another account merely named similarly
- Empty/default services (a VPC with nothing in it)
- Permission gaps reported as "secure" posture
Xalgorix Tool Strategy
terminal_executewith the AWS CLI (always available)- Pacu/CloudFox optional; every finding reproduced with raw CLI calls
- Ledger: account, principal, permissions map, resources with verified states
Specialist Handoffs
aws-iam-privilege-escalation, cloud-secrets-data-access, cloud-storage-exposure-testing, cloud-cross-account-tenant-trust, serverless-cloud-security, cloud-metadata-workload-identity (SSRF chains)
Stopping Rule
Stop when: caller identity and effective permissions are mapped, all-region resource discovery has completed (or all AccessDenied gaps are recorded), high-value secrets/data have been chased, and the strongest escalation candidates are verified or ruled out. Depth-first on live credentials and data over breadth-first listing.
