Restricted IAM key works for grants API but token endpoint rejects it

We are using an existing free US Sandbox application with an active application-bound IAM principal. It has no roles and only grants.create, grants.read and grants.delete. Its API credential is active and unexpired.

GET /v3/grants succeeds with HTTP200 using this key as a Bearer token. Hosted OAuth returns a code and our backend exchanges it within seconds using the same app ID, exact registered callback, grant_type=authorization_code and this key in client_secret. The token endpoint returns HTTP400, api.auth_failed /25022, Invalid credentials provided. No grant is created.

We also reproduced invalid_authentication / Invalid credentials provided using the documented client_credentials grant type with a deliberately nonexistent synthetic grant ID, without Google interaction. Adding the same key as a Bearer header did not change the outcome. Deliberately invalid authorization codes instead return invalid_grant, so that probe alone does not verify credential acceptance.

The IAM documentation says IAM keys may be used anywhere a Nylas API key is used, while the hosted OAuth documentation specifies an API key in client_secret.

Does POST /v3/connect/token support application-bound IAM keys in client_secret? If so, what exact grant-management permission or request format is required? Is there a known issue involving restored principal application bindings? We need a supported fix preserving the current scope and credential expiry. We cannot use an unrestricted application key, broaden roles/permissions, upgrade plans, or ask the user to repeat OAuth until this is resolved.

No credentials, codes, personal email addresses, application/principal identifiers or request identifiers are included here. Redacted diagnostics can be provided if needed through an appropriate private channel.

Hello. Our Support team has reached out to you via email to continue this investigation. Please check your mailbox for our message.