I am investigating a service account impersonation failure in a Google Cloud UAT project.
A project custom role containing only iam.serviceAccounts.getAccessToken was temporarily granted to a user on the target service account’s IAM policy. The policy used version 3 and a request.time condition limiting the grant to a 10-minute UTC window.
Observed results:
-
testIamPermissions included iam.serviceAccounts.getAccessToken.
-
A single generateAccessToken request started approximately 1.019 seconds after the successful setIamPolicy response.
-
The request used projects/-/serviceAccounts/SA_EMAIL, cloud-platform scope, a requested lifetime of 300 seconds, and no delegates.
-
It returned HTTP 403 PERMISSION_DENIED with reason IAM_PERMISSION_DENIED for iam.serviceAccounts.getAccessToken.
-
No service account token was issued.
-
The temporary binding was removed and the custom role deleted. Cleanup was verified.
The endpoint and resource format were checked. Admin Activity logs confirm the grant and cleanup, but we could not find a corresponding token-generation audit entry. The mint caller’s identity has not been independently confirmed through server logs.
IAM propagation is a hypothesis, not a confirmed cause. A Policy Troubleshooter result obtained after cleanup cannot establish authorization at the time of the failed request.
What diagnostics can distinguish propagation delay, caller-credential mismatch, condition evaluation, or another authorization restriction? What evidence should be collected before proposing a controlled retry with the same minimal permission?
We want to avoid broadening permissions or creating service account keys. Project identifiers, email addresses, credentials and internal evidence are intentionally omitted.