Resource has been temporarily blocked because we detected unusual behavior

The message indicates that Azure temporarily blocked the Foundry Models resource due to unusual or potentially abusive usage patterns. This is a service-side protection mechanism, similar in spirit to throttling (429) and other safety controls.

To address and prevent this behavior:

1. Reduce or smooth request volume
   • Avoid sudden spikes in requests.
   • Implement client-side rate limiting and exponential backoff on failures (especially around 429 or similar transient errors).
   • If using loops or parallel calls, cap concurrency so the model is not hit with bursts.
2. Check for quota and throttling conditions
   • Intermittent failures often correlate with quota or throttling conditions such as 429 errors.
   • If 429s are also observed, follow the guidance for throttling: reduce RPM/TPM usage, rebalance traffic across deployments, or request a quota increase if needed.
3. Verify request parameters are reasonable
   • Ensure max_tokens, prompt size, and tool/function payloads are not excessively large, which can look anomalous and also hit internal limits.
   • Keep payloads within documented limits for the specific model.
4. Confirm the workload complies with Responsible AI and abuse monitoring
   • Ensure prompts and usage patterns comply with Azure OpenAI / Foundry Responsible AI policies.
   • If using guardrails or blocklists, configure them appropriately so that content is filtered rather than repeatedly triggering protective blocks.
5. Monitor the deployment and utilization
   • Use Foundry monitoring to check for high utilization or error spikes.
   • If provisioned throughput (PTU) is used and utilization is near or above 100%, scale PTUs up or reduce traffic.
6. If the block persists, contact support
   • Temporary blocks should clear automatically once traffic normalizes.
   • If the message continues for an extended period even with reduced and compliant usage, open a support request through Azure Support so the service team can review logs and unblock the resource if appropriate.

References:

• Manage Azure OpenAI in Microsoft Foundry Models quota
• Monitor model deployments in Microsoft Foundry Models
• Limited access for Azure Direct Models