Test your AWS controls like you test code

Trust, but verify. A misconfigured policy can pass every audit and still let traffic through.

CATS sends real calls and real traffic from your AWS accounts, so you find the gap before anyone else does.

Get started How it works

CATS proves your controls work

  • Make the call an SCP or IAM policy should block, and check that it is denied.
  • Send HTTP, TCP and UDP traffic between accounts, across as many accounts and regions as you point it at.
  • Tag a test with the control it covers, such as SOC2-CC6.1, and the report records it.
  • Let your AI agent verify IAM, service and network behavior before it writes a test, through cats mcp or cats run.

Deploy, test, destroy. Every run leaves evidence, not infrastructure.

cats infra deploy --all
[Deploy] prod (111111111111): CREATE_IN_PROGRESS
[Deploy] dev (222222222222): CREATE_IN_PROGRESS
[Deploy] dev (222222222222): CREATE_COMPLETE in 1m 12s (14 resources)
[Deploy] dev (222222222222): Test role CATS-test-infra-dev-default-test-role authorized
[Deploy] dev (222222222222): EC2 listener i-0f3e2d1c0b9a87654 (private) ready
[Deploy] dev (222222222222): OK Created in 1m 28s, ready
[Deploy] prod (111111111111): CREATE_COMPLETE in 1m 15s (14 resources)
[Deploy] prod (111111111111): Test role CATS-test-infra-prod-default-test-role authorized
[Deploy] prod (111111111111): EC2 listener i-0a1b2c3d4e5f67890 (private) ready
[Deploy] prod (111111111111): OK Created in 1m 33s, ready
cats test
============================= test session starts ==============================
platform linux -- Python 3.12.6, pytest-9.1.1, pluggy-1.6.0
plugins: aws-cats-1.0.0
collected 4 items
 
tests/test_isolation.py::test_prod_cannot_reach_dev PASSED
tests/test_guardrails.py::test_scp_blocks_guardduty_modification PASSED
tests/test_encryption.py::test_ebs_encryption_by_default[prod_account] PASSED
tests/test_encryption.py::test_ebs_encryption_by_default[dev_account] PASSED
 
---------------------- CATS reports: run 20261031T093312Z ----------------------
4 tests (4 passed) written to /home/runner/work/controls/controls/reports
  report-20261031T093312Z.json, report.json
  assessment-results-20261031T093312Z.json, assessment-results.json
  junit-20261031T093312Z.xml, junit.xml
============================== 4 passed in 31.08s ==============================
cats infra destroy --all
[Destroy] prod (111111111111): DELETE_IN_PROGRESS
[Destroy] dev (222222222222): DELETE_IN_PROGRESS
[Destroy] dev (222222222222): OK Destroyed in 52s (14 resources)
[Destroy] prod (111111111111): OK Destroyed in 55s (14 resources)

Every account in cats.yaml is a fixture; one cats test runs the suite across all of them, on your laptop or in CI. Each run writes evidence as JSON, OSCAL, JUnit, or findings in AWS Security Hub, then the infrastructure is torn down. Reporting →

Configured isn’t the same as correct

An audit reads the policy and finds it attached. It cannot know what AWS does when the call is made. CATS makes the call from your account and records the answer.

Manual audit

Can prod turn off GuardDuty?

SCP text reviewed

CATS

Can prod turn off GuardDuty?

No. The SCP denies it.

with pytest.raises(ClientError):
  guardduty.update_detector(
    DetectorId=d, Enable=False)

PASSED AccessDeniedException: explicit deny in a service control policy

Manual audit

Can prod reach dev on 443?

Route tables reviewed

CATS

Can prod reach dev on 443?

No. The traffic is dropped.

prod.private.tcp_call(dev_host, 443)

PASSED TIMEOUT

Manual audit

Can the read-only role write to a bucket?

Policy JSON reviewed

CATS

Can the read-only role write to a bucket?

Yes. The object was written.

with pytest.raises(ClientError):
  s3.put_object(Bucket=b, Key="probe")

FAILED An inline policy on the role also grants s3:*