AWS Penetration Testing
Fail
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill details multiple pathways to achieve administrative control over an AWS environment by abusing IAM permissions.
- It describes exploiting permissions like
iam:CreateAccessKey,iam:CreateLoginProfile, andiam:AttachUserPolicyto grant the attacker full administrator access. - It provides a Python code snippet specifically designed to be injected into an AWS Lambda function (
lambda:UpdateFunctionCode) to programmatically attach theAdministratorAccesspolicy to the attacker's user identity. - [DATA_EXFILTRATION]: The skill provides methods for harvesting sensitive data and credentials from AWS infrastructure.
- It includes techniques for exploiting Server-Side Request Forgery (SSRF) to query the EC2 Instance Metadata Service (IMDSv1 and IMDSv2) and Fargate container endpoints to steal temporary IAM security credentials.
- It describes a 'Shadow Copy Attack' to extract the
NTDS.ditdatabase andSYSTEMhive from Windows Domain Controllers by mounting EBS volume snapshots on attacker-controlled instances. - It provides commands for discovering and recursively downloading the entire contents of S3 buckets using the
aws s3 synccommand. - [COMMAND_EXECUTION]: The skill instructs the agent to execute a wide range of AWS CLI and shell commands to exploit cloud resources and evade detection.
- It specifically demonstrates how to use the Systems Manager (SSM)
send-commandfunctionality to execute arbitrary shell scripts on managed EC2 instances. - It provides specific commands to delete or disable AWS CloudTrail trails to prevent logging of malicious actions.
- It suggests modifying the User-Agent string in the
Pacuframework to evade Amazon GuardDuty alerts that target common penetration testing Linux distributions. - [EXTERNAL_DOWNLOADS]: The documentation encourages downloading and executing third-party exploitation frameworks and auditing tools from various external GitHub repositories.
- Referenced tools include
Pacu(RhinoSecurityLabs),enumerate-iam(andresriancho), andaws_consoler(NetSPI). - [PERSISTENCE]: The skill explicitly outlines goals and methods for establishing persistent access within the target cloud environment for long-term red team operations.
- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect injection because it constructs shell commands using unvalidated placeholders for user-supplied data.
- Ingestion points: Variable inputs such as
TARGET_USER,ROLE_NAME,POLICY_ARN, andFUNCTION_NAMEare interpolated directly into shell commands. - Boundary markers: No delimiters or instructions are provided to the agent to treat these inputs as untrusted data or to ignore potential embedded instructions.
- Capability inventory: The skill possesses extensive capabilities including file system access, network operations (curl, git), and broad AWS API interaction via the CLI.
- Sanitization: There is no evidence of input validation, escaping, or sanitization before variable interpolation into shell execution contexts.
Recommendations
- AI detected serious security threats
Audit Metadata