Skip to content

cloud-audit

What can a hijacked AI agent reach in your AWS account, and can you prove it?

Open-source, read-only AWS security scanner. 110 checks across 25 AWS services, 31 attack-chain rules, IAM privilege escalation detection (64 methods including lateral AssumeRole graph traversal), agent-blast for Bedrock Agents and AgentCore (two threat models, OWASP Agentic and MITRE ATLAS tags), Proof Mode verification against the IAM policy simulator, data perimeter and AgentCore checks, a What-If remediation simulator, 10 Threat Feed detectors, 6 compliance frameworks, breach-cost estimation, and an MCP server for AI assistants. Every finding includes a copy-paste fix.

Quick Start

pip install cloud-audit
cloud-audit scan
cloud-audit agent-blast --verify

Try without AWS credentials:

cloud-audit agent-blast --demo
cloud-audit demo --save demo.json
cloud-audit blast-radius --report demo.json --resource arn:aws:iam::123456789012:role/support-bot-ticket-role

Why cloud-audit?

Most AWS security scanners give you a list of findings. cloud-audit goes further:

  • agent-blast - what a hijacked Bedrock Agent or AgentCore runtime, gateway or sandbox can reach: identity takeover vs behaviour takeover (prompt injection through the agent's tools), secrets, data, lateral hops, RAG poisoning, with --verify proof from the IAM policy simulator
  • Coverage gaps - a read the scanner was denied is reported as "not assessed", never as a pass
  • Attack Chains - correlates individual findings into exploitable attack paths (31 rules based on MITRE ATT&CK)
  • Root Cause Grouping - "fix 4 things, break 22 chains" prioritization
  • What-If Simulator - cloud-audit simulate --fix aws-vpc-002 shows impact before you change anything
  • IAM Privilege Escalation - 64 escalation methods (PassRole, policy mutation, permission boundary bypass, resource policy abuse, compute hijack, SSM credential theft, and lateral movement via AssumeRole trust-graph traversal)
  • Proof Mode - cloud-audit scan --verify confirms escalation paths against the IAM policy simulator (read-only)
  • Blast Radius - walk outward from any resource to see what an attacker reaches; export to the live visualizer
  • Threat Feed - 10 detectors for active-abuse patterns from 2025-2026 incidents
  • Data Perimeter - resource-policy confused-deputy and cross-org exposure, evaluating condition values
  • AgentCore Security - Amazon Bedrock AgentCore AI agent checks (network mode, MMDSv2, gateway authorizer)
  • AI-SPM - Bedrock model invocation logging, guardrails, SageMaker notebook security
  • Security Trends - cloud-audit trend tracks score, chains, and risk over time
  • Compliance Engine - maps findings to 6 frameworks (CIS AWS v3.0, SOC 2 + 4 Beta) with per-control evidence
  • Breach Cost Estimation - dollar ranges on security findings based on IBM/Verizon breach data
  • Remediation on every finding - AWS CLI commands and Terraform HCL you can copy-paste
  • MCP Server - ask Claude Code or Cursor to scan your AWS account

Who Uses This

  • Small teams without a security team - attack chains show you which findings actually matter
  • Consultants auditing client accounts - generate a professional report in one command: cloud-audit scan --format html -o report.html
  • DevOps/SRE running pre-deploy checks - exit codes + SARIF for CI/CD gating
  • Teams preparing for compliance audits - 6 frameworks including CIS AWS v3.0, SOC 2, BSI C5, ISO 27001, HIPAA, and NIS2