One stolen token: tracing blast radius in a GovCloud account
Zero trust starts with a simple exercise. Pick one credential, assume it leaked, and follow everywhere it can reach.
Zero trust strategies tend to start with pillars and maturity levels. We like to start with a smaller exercise that produces a concrete answer in an afternoon: pick one credential, assume an adversary has it, and trace everything it can reach.
The answer is the blast radius of that credential. Shrinking it is most of what zero trust means in practice.
Pick a realistic credential
Do not start with the root account. Start with something that leaks in real incidents:
- An access key for a CI/CD service role, stored as a pipeline variable.
- A developer’s federated session token, cached on a laptop.
- The instance role attached to an internet-facing workload.
Each of these regularly ends up somewhere it should not be: a build log, a compromised workstation, a server-side request forgery bug.
Trace what it can do directly
For the chosen principal, list its effective permissions. In AWS that means the identity policies, any permission boundary, session policies, and service control policies above the account. IAM Access Analyzer and the policy simulator help, but read the policies yourself too. Wildcards hide in places tools summarize away.
Write the answer as actions on resources, not policy names: “can read every object in these four buckets”, “can pass any role to Lambda”, “can modify security groups in this VPC”.
Trace what it can become
This is where blast radius usually explodes. Look for paths from the starting principal to more privilege:
- Role assumption. Which roles trust this principal, directly or through an overly broad trust policy?
- Pass role. Can it hand a more powerful role to a compute service it can also create or update?
- Resource policies. Do bucket, key or queue policies grant access to this principal outside its identity policy?
- Secrets. Can it read parameters or secrets that contain other credentials?
- Compute it can modify. If it can change a function’s code or an instance’s user data, it holds that compute’s role too.
Follow each path until it stops. Then repeat from the new principal. The full graph is the real blast radius.
Trace where it can reach on the network
Identity is half the picture. If the credential can modify security groups, route tables or VPC endpoint policies, it can open network paths to systems its permissions never mention. Note any path from the starting workload into enclaves that hold mission data.
Shrink it
With the graph drawn, the fixes are usually clear and rarely expensive:
- Replace static keys with short-lived credentials through federation or workload identity.
- Scope trust policies to specific principals and add conditions on source and session tags.
- Remove
iam:PassRoleon wildcards; name the exact roles. - Move secrets behind resource policies that name their single consumer.
- Add detection on the steps an attacker must take: unusual role assumption, policy changes, new access keys.
Then run the exercise again from the same starting point. If the graph is smaller, you made measurable progress against the DoD Zero Trust Reference Architecture’s identity and device pillars, and you can show it.
Make it routine
We run this exercise at the start of every cloud engagement because it turns an abstract maturity model into a list of specific changes. You can run it yourself quarterly on your highest-risk credentials. The first time takes a day. After that, the difference between this quarter’s graph and last quarter’s is the most honest zero trust metric you will have.