Azure AI Foundry HTTP 403 "unusual behavior" block on Elevate-grant resource

That 403 sounds like a service-side protection or abuse-prevention block rather than a normal RBAC permission problem. Creating more retries usually makes this worse, because the resource may continue to look suspicious to the protection system.

I would stop the traffic from that resource, collect the subscription id, resource name, region, timestamps, request ids, and a short explanation of the workload, then open an Azure support case against Azure AI Foundry or Azure OpenAI. If the workload is legitimate, support is the right path to review and clear the block. In parallel, check for accidental loops, very high request rates, or test scripts that may have generated traffic patterns the service considered unusual.