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, 64 IAM privilege-escalation methods, 31 attack-chain rules, blast radius for any resource, and agent-blast for Bedrock Agents and AgentCore: what a prompt-injected or credential-stolen agent can reach, with the IAM policy simulator as the witness. Every finding ships an AWS CLI + Terraform fix. Nothing is written to your account.

30-second demo - agent-blast - Full scan - What's inside - Proof Mode - Installation - Documentation

## Thirty seconds, no AWS account needed ```bash pip install cloud-audit cloud-audit agent-blast --demo # a hijacked Bedrock Agent, two threat models, one screen cloud-audit demo --save demo.json # a full sample scan, then explore it offline: cloud-audit blast-radius --report demo.json --resource arn:aws:iam::123456789012:role/support-bot-ticket-role cloud-audit simulate --report demo.json --fix aws-iam-018 cloud-audit exposure --report demo.json ``` The sample account is invented. Everything derived from it (attack chains, root causes, breach cost, security graph, agent reach) is produced by the same engines a real scan uses. With credentials, the real thing is one command and read-only. The AWS-managed `SecurityAudit` policy covers every check ([permissions](#aws-permissions)): ```bash cloud-audit scan # default profile and region cloud-audit scan --verify # plus IAM policy-simulator proof for escalation paths cloud-audit agent-blast --verify # then: what your agents can reach, simulator-confirmed ``` ## agent-blast: what a hijacked AI agent can reach An AI agent in AWS is a bundle of IAM identities: the role the agent runs as, the execution roles of the Lambda functions behind its tools, the roles of its knowledge bases, gateways and sandboxes. When the agent is hijacked, the attacker acts with *those* identities. `agent-blast` answers what that means, per agent, under two threat models: | Threat model | What the attacker has | What is in reach | |---|---|---| | **Identity takeover** | the credentials of a role the agent runs as (sandbox escape, metadata-service read, leaked session) | the full role: privilege-escalation methods, AssumeRole hops, data | | **Behaviour takeover** | nothing but a prompt (indirect prompt injection through a document, a ticket, a web page) | what the agent's tools can do, bounded by the tools' own execution roles | Real output for the sample agent (`cloud-audit agent-blast --demo -a support-bot`): ``` support-bot (bedrock_agent, eu-central-1) ├── A hijacked tool (tool: create-ticket) can escalate to account admin via PassRole+Lambda ├── status: PREPARED ├── Identity takeover (attacker holds the role's credentials) │ └── support-bot-agent-role arn:aws:iam::123456789012:role/support-bot-agent-role │ └── no escalation path; 3 reach(es) from policy grants, 1 on named resources (see Reach) │ (risk 0/100, 1 nodes) ├── Behaviour takeover (prompt injection: the agent's tools, their roles) │ ├── search-docs action_group_lambda -> support-bot-search-role │ │ ├── 3 OpenAPI path(s) in apiSchema; state ENABLED │ │ └── no escalation path; 4 reach(es) from policy grants, 4 on named resources (see Reach) │ ├── create-ticket action_group_lambda -> support-bot-ticket-role │ │ ├── 2 function(s) in functionSchema; state ENABLED │ │ ├── reaches Account Takeover in 1 hop(s) via PassRole+Lambda (risk 72/100) │ │ └── escalation: PassRole+Lambda │ └── product-docs knowledge_base -> support-bot-kb-role │ ├── 1 S3 data source(s) │ └── no escalation path; 2 reach(es) from policy grants, 2 on named resources (see Reach) ├── Reach (13 action/resource pairs) │ ├── tool: search-docs secretsmanager:GetSecretValue on │ │ arn:aws:secretsmanager:eu-central-1:123456789012:secret:prod/db-credentials-Ab12Cd │ │ unverified read secret values │ ├── tool: create-ticket sts:AssumeRole on arn:aws:iam::123456789012:role/ops-admin unverified │ │ assume IAM roles │ ├── tool: create-ticket lambda:InvokeFunction on * unverified invoke Lambda functions │ ├── tool: search-docs s3:PutObject on arn:aws:s3:::company-backups-2024/* unverified write S3 │ │ objects │ ├── support-bot-agent-role s3:GetObject on arn:aws:s3:::kb-product-docs/* unverified read the │ │ agent's knowledge base │ ├── tool: search-docs s3:GetObject on arn:aws:s3:::company-backups-2024/* unverified read S3 │ │ objects │ ... ├── Tags │ ├── OWASP Agentic: ASI02 Tool Misuse and Exploitation, ASI03 Identity and Privilege Abuse, │ │ ASI05 Unexpected Code Execution │ └── MITRE ATLAS: AML.T0034 Cost Harvesting, AML.T0040 AI Model Inference API Access, AML.T0053 │ AI Agent Tool Invocation, AML.T0086 Exfiltration via AI Agent Tool Invocation └── Simulated, not executed: identity policies, the attached permissions boundary and SCPs are evaluated by the IAM policy simulator; resource-based policies of the targets, RCPs and VPC endpoint policies are not. Tool reach is an upper bound set by the tool's role; what the tool code does with its inputs is narrower. ``` The story the output tells: the support bot itself is harmless, but a document that tricks it into calling `create-ticket` runs code under a role that can pass an admin role to a new Lambda. The `search-docs` tool, meant to read product docs, can also read the production database secret and write into the backups bucket. Every line names the identity, the action and the resource, so the fix is a policy statement, not a debate. The same engine covers AgentCore. For the sample gateway (`--demo -a tools-gw`) the headline is a cross-agent one: ``` tools-gw (agentcore_gateway, eu-central-1) ├── tool: crm-sync can write into agent 'support-bot' knowledge base (RAG poisoning): arn:aws:s3:::kb-product-docs/* ├── inbound authorizer: NONE; policy engine: none ... │ ├── tool: crm-sync s3:PutObject on arn:aws:s3:::kb-product-docs/* unverified write into agent │ │ 'support-bot' knowledge base (RAG poisoning) ... ├── Coverage notes │ └── tool 'erp-mcp' (gateway_target_mcp_server): backend identity unknown, not assessed - remote │ MCP server; credential provider: OAUTH ``` What `agent-blast` reads (all list/get, no charge): - **Bedrock Agents**: the agent resource role, every action group at the `DRAFT` version resolved to its Lambda and the Lambda's execution role, knowledge bases resolved to their S3 data-source buckets. - **AgentCore**: runtime role and network mode, gateway role with authorizer, policy engine and every target (Lambda, remote MCP server, API Gateway, OpenAPI, Smithy) with its credential provider, code interpreter and browser execution roles. - **IAM policies** of exactly those identities (inline, attached, group), kept raw: action, resource, effect, whether a `Condition` is present. Conditions are flagged, never guessed. The simulator decides. Then, per identity: the 64 privilege-escalation methods, AssumeRole hops through the trust graph, and every data, secret, lateral, code-execution and model-invocation grant matched against concrete resources. A knowledge-base bucket is always a concrete target, even behind `s3:*` on `*`, and write access to it is reported as RAG poisoning. Anything the scanner could not read (a denied region, a tool without an IAM identity) is a **coverage note**, never a silent pass. `--verify` asks `iam:SimulatePrincipalPolicy` about every concrete pair, with the context an attacker really has (no MFA on the session, TLS transport), and marks each reach `PROVEN`, `DENIED` (with the policy layer that denied: SCP, permissions boundary) or `not asserted`. Output formats: `tree`, `json`, `markdown`. The MCP server exposes the same report as `get_agent_blast`. ## A full scan `cloud-audit scan` runs 110 read-only checks across 25 AWS services, correlates the findings into attack chains, ranks the fixes by how many chains they break and prices the exposure. This is the sample account (`cloud-audit demo`), rendered by the same code as a real scan and trimmed for length: ``` ┌─────── Health Score ───────┐ │ 15 / 100 │ └────────────────────────────┘ Provider AWS Account 123456789012 Regions eu-central-1 Resources scanned 80 Checks passed 14 Checks failed 6 Attack chains 3 Risk exposure $572K - $12.5M (IBM/Verizon data) ┌───────────────────────── Attack Chains (3 detected) ─────────────────────────┐ │ CRITICAL PassRole Escalation to Admin (support-bot-ticket-role) │ │ Principal 'support-bot-ticket-role' has iam:PassRole and │ │ can create resources with privileged roles via: │ │ PassRole+Lambda. The attacker passes an admin role to a │ │ service, then uses that service to execute code with admin │ │ privileges. │ │ Fix: Scope iam:PassRole for 'support-bot-ticket-role' to │ │ specific role ARNs │ │ Risk: $250K - $5.0M │ │ │ │ CRITICAL Internet-Exposed Admin Instance │ │ i-0abc123def456789, sg-0a1b2c3d4e5f67890 │ │ Instance i-0abc123def456789 is reachable from the internet │ │ via open security group and has admin IAM role │ │ 'prod-admin-role'. An attacker can reach the instance, │ │ access IMDS credentials, and gain full admin access to the │ │ AWS account. │ │ Fix: Restrict security group sg-0a1b2c3d4e5f67890 to │ │ specific IPs (effort: LOW). │ │ Risk: $62K - $625K │ │ │ │ CRITICAL CI/CD to Admin Takeover │ │ Role 'github-deploy-role' trusts an OIDC provider without │ │ restricting the 'sub' claim AND has admin permissions. Any │ │ repository on the CI/CD platform can assume this role and │ │ gain full AWS admin access. │ │ Fix: Add 'sub' condition to the trust policy (effort: │ │ LOW). │ │ Risk: $125K - $1.2M │ └──────────────────────────────────────────────────────────────────────────────┘ ┌────────────── Remediation Plan (fix 3 things, break 3 chains) ───────────────┐ │ # Fix Effort Chains Findings Risk Red │ │ 1 Restrict security groups LOW 1 1 $62.5K - │ │ 2 Add OIDC 'sub' condition to LOW 1 1 $125K - │ │ trust policies │ │ 3 Remove IAM privilege MEDIUM 1 1 $250K - │ │ escalation paths │ └──────────────────────────────────────────────────────────────────────────────┘ ┌─────────────────── Quick Wins (break CRITICAL chains now) ───────────────────┐ │ 1. aws ec2 revoke-security-group-ingress --group-id sg-0a1b2c3d4e5f67890 │ │ --protocol tcp --port 22 --cidr 0.0.0.0/0 │ │ Breaks: AC-01 (Internet-Exposed Admin Instance) │ │ │ │ 2. aws iam update-assume-role-policy --role-name github-deploy-role │ │ --policy-document file://trust-with-sub.json │ │ Breaks: AC-07 (CI/CD to Admin Takeover) │ └──────────────────────────────────────────────────────────────────────────────┘ ``` Preview a fix before touching anything. `simulate` removes the findings a check would fix and recomputes score, chains and exposure offline: ``` $ cloud-audit simulate --report demo.json --fix aws-iam-018 Simulation: Apply Remove IAM privilege escalation paths ┌────────────────────────────── Impact Analysis ───────────────────────────────┐ │ Metric Before After Delta │ │ Health Score 15/100 35/100 +20 │ │ Attack Chains 3 2 -1 │ │ Findings 6 5 -1 │ │ Risk Exposure $572K - $12.5M $187.5K - $1.9M -85% │ └──────────────────────────────────────────────────────────────────────────────┘ Chains broken (1): CRITICAL PassRole Escalation to Admin (support-bot-ticket-role) -> GONE Chains remaining (2): CRITICAL Internet-Exposed Admin Instance CRITICAL CI/CD to Admin Takeover Next recommended fix: Restrict security groups -> breaks 1 more chain(s) ``` Walk outward from any resource to see what its compromise reaches: ``` $ cloud-audit blast-radius --report demo.json --resource arn:aws:iam::123456789012:role/support-bot-ticket-role ┌────────────────────────────────────── Blast Radius ──────────────────────────────────────┐ │ Compromise of role/support-bot-ticket-role leads to Account Takeover via 2 hop(s). │ │ Resource: arn:aws:iam::123456789012:role/support-bot-ticket-role │ │ Reachable: 3 | Paths to impact: 1 | Risk score: 72/100 │ └──────────────────────────────────────────────────────────────────────────────────────────┘ role/support-bot-ticket-role - IAM Role (assumed compromised) ├── Account Takeover - Full AWS admin (via grants admin, type=exploit) └── arn:aws:s3:::company-backups-2024 - S3 bucket (reachable via admin) (via s3:GetObject + ListBucket, type=data) Available fixes (from scan findings): - aws-iam-018 Role 'support-bot-ticket-role' can escalate via PassRole+Lambda (breaks chain, effort=MEDIUM) - aws-s3-001 Bucket 'company-backups-2024' has Block Public Access disabled (breaks chain, effort=LOW) ``` `--format json` drops straight into the open browser visualizer at [blast-audit.haitmg.pl](https://blast-audit.haitmg.pl/); `--format mermaid` gives a diagram for a wiki page. `exposure` ranks every node of the security graph by internet reachability and escalation: ``` $ cloud-audit exposure --report demo.json -n 6 Score Type Label Why 60 iam_role prod-admin-role internet-reachable 41 internet Internet internet-reachable 41 ec2 i-0abc123def456789 internet-reachable 41 security_group sg-0a1b2c3d4e5f67890 internet-reachable 20 iam_role support-bot-ticket-role escalation 20 iam_role support-bot-search-role linked to others ```

_{The visualizer runs entirely in your browser on a blast-radius --format json export.}

## What's inside | Capability | Size | Command | |---|---|---| | Configuration checks, each with an AWS CLI and Terraform fix | **110 checks, 25 services**: IAM, S3, EC2, VPC, RDS, KMS, CloudTrail, CloudWatch, GuardDuty, Config, Lambda, ECS, EFS, SSM, Secrets Manager, DynamoDB, Backup, Inspector, Security Hub, WAF, Bedrock, SageMaker, Bedrock AgentCore, Data Perimeter, account | `scan`, `list-checks` | | IAM privilege escalation | **64 methods in 9 categories**: IAM self-mutation, credential access, PassRole to a service, Lambda code modification, trust-policy abuse, permission-boundary bypass, resource-policy abuse, compute hijack, lateral AssumeRole (BFS over the trust graph) | `scan` (`aws-iam-018`) | | Attack chains | **31 rules** correlating findings and live relationships (EC2 to role, Lambda to role, OIDC to policy) into paths with MITRE ATT&CK references and a priority fix | `scan` | | Root causes and quick wins | groups findings by the fix that removes them and ranks by chains broken | `scan` | | What-If | before/after score, chains and exposure for any set of checks, offline | `simulate --fix` | | Blast radius | forward BFS from an EC2 instance, IAM role or user, Lambda, S3 bucket or secret; JSON contract for the visualizer | `blast-radius` | | Exposure ranking | every security-graph node scored by internet reachability and escalation | `exposure` | | AI agents | inventory of Bedrock Agents and AgentCore runtimes, gateways, code interpreters, browsers with their IAM identities and tools; two-threat-model reach; OWASP Agentic and MITRE ATLAS tags; simulator proof | `agent-blast` | | Proof Mode | `iam:SimulatePrincipalPolicy` on escalation paths (broad) and on concrete (principal, action, resource) triples (agent-blast) | `scan --verify`, `agent-blast --verify` | | Threat feed | **10 detectors** of active abuse, each tied to a documented 2025-2026 incident pattern, versioned as a rules pack | `threat-feed` | | Data perimeter | 5 checks on resource policies: confused deputy and cross-organization exposure, evaluating condition values | `scan` (`aws-dp-*`) | | AgentCore configuration | 6 checks: code interpreter and runtime network mode, MMDSv2, memory KMS, gateway authorizer, gateway policy engine | `scan` (`aws-agc-*`) | | Breach cost | USD range per finding and per chain with the public source behind each number | `scan` | | Compliance | **6 frameworks** with per-control evidence: CIS AWS v3.0, SOC 2 Type II, ISO 27001:2022, HIPAA, NIS2, BSI C5:2020 | `scan --compliance`, `list-frameworks` | | Drift and trend | diff two scans with exit codes; posture history with sparklines | `diff`, `trend` | | Coverage gaps | a read the scanner was denied is reported as "not assessed", never as a pass | every command | | Outputs | console, JSON, HTML report, SARIF (GitHub Code Scanning), Markdown | `scan --format` | | MCP server | 7 read-only tools for Claude Code, Cursor and any MCP client, including `get_agent_blast` | `cloud-audit-mcp` | The ten threat-feed detectors, as `cloud-audit threat-feed --list` prints them (rules pack `2026-Q2`): | Pattern | Severity | What it catches | |---|---|---| | TF-001 | MEDIUM | SES recently verified identity (phishing setup precursor) | | TF-002 | HIGH | Public Lambda Function URL (`AuthType=NONE`) | | TF-003 | CRITICAL | AWS-attached credential quarantine policy | | TF-004 | CRITICAL | Leaked-credentials scanner user-agent observed in CloudTrail | | TF-005 | HIGH | Recently-created IAM role with cryptomining-style compute access | | TF-006 | HIGH | EC2 or AgentCore using IMDSv1 / MMDSv1 (SSRF credential theft vector) | | TF-007 | MEDIUM | CI/CD role at risk of whoAMI confusion attack | | TF-008 | HIGH | CloudTrail tampering precursor (logging stopped or delivery failing) | | TF-009 | HIGH | IAM Roles Anywhere trust anchor with external CA | | TF-010 | HIGH | `AmazonDataZoneFullAccess` attached to a non-admin principal | ## Proof Mode: simulated, not guessed Static analysis of IAM policies over-reports. Proof Mode asks AWS's own authorization engine, `iam:SimulatePrincipalPolicy`, which is read-only and carries no per-call charge. - `scan --verify` checks every detected escalation path. `verified: true` means the simulator allowed every required action; `false` means it denied one (the path is very likely a false positive); `null` means not asserted, with the reason in `verification_detail`. - `agent-blast --verify` checks every concrete (principal, action, resource) reach, supplies the context an attacker with stolen or assumed credentials has, and names the layer that denied. What the simulator evaluates as of 2026-07-30: identity policies, the attached permissions boundary, the organization's SCPs including their condition keys, and the context keys you supply. What it does not: resource control policies, VPC endpoint policies, role chaining, and resource-based policies for IAM roles. cloud-audit says so on every verified line. A `true` proves the permission exists. It does not prove an end-to-end attack; nothing is executed. ## Reports ```bash cloud-audit scan --format html -o report.html # client-ready cloud-audit scan --format sarif -o results.sarif # GitHub Code Scanning cloud-audit scan --format json -o report.json # machine-readable, feeds every offline command cloud-audit scan --format markdown -o report.md # PR comments cloud-audit scan --compliance cis_aws_v3 --format html -o cis.html cloud-audit agent-blast --format markdown -o agents.md ``` ## CI/CD ```yaml - run: pip install cloud-audit - run: cloud-audit scan --format sarif --output results.sarif - uses: github/codeql-action/upload-sarif@v3 with: sarif_file: results.sarif ``` `--quiet` exits with a code only: `0` clean, `1` findings, `2` error. Gate on severity with `--min-severity high`. `cloud-audit diff old.json new.json --quiet` exits `1` on new findings, so a daily scan catches ClickOps drift. The repository is also a composite GitHub Action (`gebalamariusz/cloud-audit@v0`). Ready-made workflows: [basic scan](examples/github-actions.yml), [daily diff](examples/daily-scan-with-diff.yml), [post-deploy](examples/post-deploy-scan.yml). ## Installation ```bash pip install cloud-audit # pip (recommended) pipx install cloud-audit # isolated uvx --from cloud-audit cloud-audit scan # no install docker run ghcr.io/gebalamariusz/cloud-audit scan # Docker (also on Docker Hub: haitmg/cloud-audit) ``` Docker with credentials: ```bash docker run -v ~/.aws:/home/cloudaudit/.aws:ro ghcr.io/gebalamariusz/cloud-audit scan ``` Python 3.10 to 3.13. Dependencies: boto3, typer, rich, pydantic, jinja2, pyyaml, mcp. ## AWS permissions Read-only. Attach the AWS-managed `SecurityAudit` policy; it covers every check, the IAM escalation analysis, the agent inventory and Proof Mode: ```bash aws iam attach-role-policy --role-name auditor \ --policy-arn arn:aws:iam::aws:policy/SecurityAudit ``` cloud-audit never modifies your infrastructure. `simulate`, `blast-radius`, `exposure`, `diff`, `trend` and `agent-blast` run locally against a saved scan and make no AWS calls; `--verify` adds simulator calls only. Regions where a service does not exist are skipped; regions where the scanner is denied a read are listed as coverage gaps. ## MCP server Seven read-only tools for AI assistants: `scan_aws`, `get_findings`, `get_attack_chains`, `get_remediation`, `get_health_score`, `list_checks`, `get_agent_blast`. ```bash claude mcp add cloud-audit -- uvx --from cloud-audit cloud-audit-mcp ``` Then ask: *"scan my AWS account and tell me what my Bedrock agents could reach if hijacked"*.

Common flags and configuration

```bash cloud-audit scan -R # show remediation inline cloud-audit scan --profile prod --regions eu-central-1 # profile / region cloud-audit scan --regions all # all enabled regions cloud-audit scan --role-arn arn:aws:iam::...:role/audit # cross-account cloud-audit scan --export-fixes fixes.sh # every CLI fix as a script cloud-audit scan --categories security,cost # skip reliability/performance checks cloud-audit agent-blast -a support-bot --profile prod --verify ``` Defaults live in `.cloud-audit.yml` (regions, `min_severity`, `exclude_checks`, time-boxed `suppressions`). Environment variables (`CLOUD_AUDIT_REGIONS`, `CLOUD_AUDIT_MIN_SEVERITY`, ...) override the file; CLI flags override everything. See the [configuration guide](https://haitmg.pl/cloud-audit/configuration/config-file/).

## Documentation Full documentation at **[haitmg.pl/cloud-audit](https://haitmg.pl/cloud-audit/)**: [getting started](https://haitmg.pl/cloud-audit/getting-started/installation/), [agent-blast](https://haitmg.pl/cloud-audit/features/agent-blast/), [attack chains](https://haitmg.pl/cloud-audit/features/attack-chains/), [IAM escalation](https://haitmg.pl/cloud-audit/features/iam-escalation/), [blast radius](https://haitmg.pl/cloud-audit/features/blast-radius/), [Proof Mode](https://haitmg.pl/cloud-audit/features/proof-mode/), [threat feed](https://haitmg.pl/cloud-audit/features/threat-feed/), [data perimeter](https://haitmg.pl/cloud-audit/features/data-perimeter/), [AgentCore](https://haitmg.pl/cloud-audit/features/agentcore/), [compliance](https://haitmg.pl/cloud-audit/compliance/overview/), and the [full check reference](https://haitmg.pl/cloud-audit/checks/). ## Commercial support cloud-audit is free and stays free. If you want a human on the findings, the author offers professional services: - **Scanner output review (free)**: send your cloud-audit, Prowler or Security Hub output, get a short written review of what actually matters and what to fix first - **AWS security audit**: full account audit with a prioritized report and ready-to-apply fixes - **AI agent access review**: what your Bedrock and AgentCore agents can reach, with the policy changes to narrow it - **Remediation support**: Terraform and IAM changes, verified against your workloads - **Palo Alto VM-Series on AWS**: architecture and security review (GWLB/TGW, HA, routing) Details: [haitmg.pl/cloud-audit-support](https://haitmg.pl/cloud-audit-support/?utm_source=github&utm_medium=readme) or email [kontakt@haitmg.pl](mailto:kontakt@haitmg.pl). ## Development ```bash git clone https://github.com/gebalamariusz/cloud-audit.git cd cloud-audit pip install -e ".[dev]" pytest -q && ruff check src/ tests/ && mypy src/ ``` See [CONTRIBUTING.md](CONTRIBUTING.md) to add a check. Releases in [CHANGELOG.md](CHANGELOG.md). ## License [MIT](LICENSE) - Mariusz Gebala / [HAIT](https://haitmg.pl)