Skip to content

Free tools

AWS Policy Checker

Paste an IAM identity policy, a role trust policy or a resource policy such as an S3 bucket policy. The page says in plain words what it allows and flags the grants an attacker looks for first, each with the statement it came from and what to change.

No network. Nothing you enter leaves this tab. The policy is parsed and checked in your browser, nothing is stored, and nothing about what you paste is counted or logged.

Load a sample

One policy document as JSON, up to 100 KB. Output from aws iam get-role, get-policy-version or aws s3api get-bucket-policy works too: the page reads the policy inside it. Ctrl or Cmd and Enter checks it.

The 12-digit account the policy belongs to, so a principal in that account is not reported as another account. It stays in the tab like the policy.

Results

What the policy allows, each finding and where it came from appear here.

Reading the results

Full administrator and whole services
Action * on Resource * is administrator. A whole service such as s3:* or sts:* is close to it for that service, and iam:* is administrator in all but name, because it can grant itself the rest.
Allow with NotAction or NotResource
Both invert the list: every action, or every resource, except the ones named, including ones added later. With Deny that is useful; with Allow it is nearly always broader than the author meant.
Public principals
Principal * or {"AWS": "*"} admits every AWS account and, for services such as S3, the whole internet. Only a condition on who, such as aws:PrincipalOrgID, aws:SourceVpce or aws:SourceAccount with a positive operator, counts as limiting it.
Cross-account and web identity trust
A trust policy that names another account lets that account's administrators decide who becomes your role. Vendors need sts:ExternalId and your own accounts aws:PrincipalOrgID. A GitHub OIDC trust with no sub condition admits every repository on GitHub.
iam:PassRole and what it pairs with
PassRole runs nothing on its own. Paired with lambda:CreateFunction and lambda:InvokeFunction, ec2:RunInstances, cloudformation:CreateStack or glue:CreateDevEndpoint, it runs code as the role it passes, and on every role that can be the strongest one.
Escalation actions
Policy versions, attaching and writing policies, group membership and trust rewrites change permissions; access keys and login profiles take over other users; lambda:UpdateFunctionCode, ssm:SendCommand and ssm:StartSession run code as a stronger role.
Broad data reads
secretsmanager:GetSecretValue, ssm:GetParameter*, kms:Decrypt and s3:GetObject on every resource read whatever the account holds, not only what the workload needs.
Conditions and Deny
Only a condition that narrows who or where, with a literal value such as an account, a network or a role, lowers a finding one step; a wildcard value, a negated operator or a Bool such as MFA is named but lowers nothing. A Deny is listed and never counted as a grant, and a Deny with no condition on every resource cancels the grant it covers.

What this check cannot see

A policy document is one input to an AWS decision. This page reads only the document you paste, so it cannot see:

  • Service control policies and resource control policies from AWS Organizations, which can deny what this policy allows.
  • Permissions boundaries, which cap what an identity policy can grant.
  • Session policies passed when a role is assumed, which narrow that one session.
  • The resource's own policy (a bucket, key, queue or secret policy), which can deny the request or, inside one account, grant it on its own. For a resource policy, the caller's identity policies are the part it cannot see.
  • Other policies attached to the same user, group or role, which add to what this one grants.
  • S3 Block Public Access settings and bucket or object ACLs, which can stop a public grant or open access that no policy shows.
  • VPC endpoint policies, which can narrow requests that pass through an endpoint.
  • KMS grants, which give use of a key outside any key policy.
  • Whether the resources, roles and accounts named here exist, and what they hold.
  • The values conditions test at request time, such as tags, source addresses and MFA.

A clean result means none of the patterns this page knows appear in the policy. It is not proof the policy is safe: what it can really do depends on everything above, and on the IAM policy simulator or a test run against the account itself.