Atlas

AWS Cloud Adversary Emulation Platform

⚠️ This tool is still under development. APIs and behavior may change. Use with caution in production environments.

Attack Knowledge: pathfinding.cloud & Hacking the Cloud

Atlas incorporates all techniques from pathfinding.cloud (parginfd.cloud — Datadog's verified AWS IAM privilege escalation paths). The 65+ attack paths—covering IAM, EC2, Lambda, ECS, Glue, CodeBuild, SageMaker, SSM, CloudFormation, Bedrock, AppRunner, and more—are synced automatically on first run and merged into the attack graph.

Atlas also includes techniques from Hacking the Cloud — exploitation paths (IMDS credential theft, Lambda SSRF, misconfigured resource policies, OIDC trust misconfig, Cognito self-signup, CloudFront/S3 takeover, PassRole abuse), post-exploitation persistence (GetFederationToken survival, CodeBuild runner persistence, rogue OIDC provider, IAM Roles Anywhere, S3 ACL persistence), and defense evasion (GuardDuty, CloudTrail). Operational tradecraft from Hacking the Cloud is in post_exploitation_operational.yaml.

This gives Atlas comprehensive, battle-tested coverage of real-world AWS privilege escalation techniques.

Contributing Red Team Techniques

We welcome contributions of new red team techniques. If you have attack paths, privilege escalation methods, or AWS abuse techniques you'd like to add to Atlas, please open an issue or submit a pull request. The planner and attack graph are designed to be extended—see src/atlas/planner/attack_graph.py and src/atlas/knowledge/data/api_detection_profiles.yaml for how techniques are modeled.

What is Atlas?

Atlas is a next-generation AWS cloud adversary emulation platform. It helps red teams and security researchers:

• Discover attack paths from a given identity (recon + attack graph)
• Plan multi-step privilege escalation chains
• Simulate execution without making AWS API calls (BloodHound-style — mapping only, no execution)
• Explain attack paths with AI-powered or template-based explanations

Features

Requirements

• Python 3.12+
• AWS credentials configured (e.g. ~/.aws/credentials)

Installation

Install from PyPI (recommended):

pip install atlas-redteam

Or with pipx (isolated environment, no venv needed):

pipx install atlas-redteam

Update to latest version:

pip install --upgrade atlas-redteam
# or
pipx upgrade atlas-redteam

Note for maintainers: To publish new versions so users get updates, see docs/RELEASE.md.

For development (editable install):

git clone https://github.com/Haggag-22/Atlas.git
cd Atlas
pip install -e ".[dev]"

Quick Start

# Configure AWS profile
atlas config --profile my-profile --region us-east-1

# Run recon + planning (creates a case)
atlas plan --case mycase

# List attack paths and simulate
atlas simulate --case mycase --attack-path AP-01

# Explain an attack path
atlas explain --case mycase --attack-path AP-01

# Open the GUI (k8scout-style, port 8050)
atlas gui --case mycase

Note: If you see the old Streamlit UI (port 8501), you may have run streamlit run directly. Use atlas gui instead for the new D3.js-based interface.

Commands

Output Structure

output/<case>/
├── case.json           # Case metadata
├── plan/               # Recon + planning
│   ├── env_model.json
│   ├── attack_edges.json
│   ├── graph.json
│   ├── attack_paths.json
│   └── ...
├── sim/                # Simulation results (if run)
└── explanations.json   # Cached AI/template explanations

Attack Techniques

Atlas models techniques from pathfinding.cloud, Hacking the Cloud, and its own attack pattern registry, including:

• IAM: Role assumption, access key creation, policy attachment, inline policy injection, PassRole abuse, trust policy modification, permissions boundary manipulation
• EC2: PassRole via instance profile, userdata read/modify, EC2 Instance Connect, IMDS credential theft (Hacking the Cloud)
• Lambda: PassRole, code/config injection, credential theft, SSRF-based cred theft (Hacking the Cloud)
• ECS, Glue, CodeBuild, SageMaker: PassRole and service-specific escalation
• SSM: Session Manager, tag-based enablement
• CloudFormation, Bedrock, AppRunner: PassRole and resource abuse
• S3: Bucket policy, object read/write, ACL persistence (Hacking the Cloud)
• Cognito, OIDC, CloudFront: Self-signup, trust misconfig, domain takeover (Hacking the Cloud)
• Post-exploitation: GetFederationToken survival, rogue OIDC provider, IAM Roles Anywhere (Hacking the Cloud)
• Defense evasion: GuardDuty, CloudTrail evasion (Hacking the Cloud)

Detection costs and noise levels are derived from CloudTrail and GuardDuty profiles in src/atlas/knowledge/.

Discovered Resources

The recon engine collects the following resource types (configurable via recon.resource_types):

License

MIT