Troubleshoot common HTTP errors - Azure OpenAI in Microsoft Foundry

Learn how to diagnose and resolve common HTTP errors when using Azure OpenAI in Microsoft Foundry Models, including 400, 401, 403, 404, 429, and 500 errors.

This article helps you diagnose and resolve common HTTP errors returned by Azure OpenAI in Microsoft Foundry Models inference endpoints.

Error reference

HTTP Status	Error	Common Causes	Resolution
400	Bad Request	Invalid request payload, context length exceeded, unsupported parameters, content filtering blocked a prompt or response	Check the error message for the validation failure. Content filtering returns HTTP 400 with the content_filter code. See Content filtering.
401	Unauthorized	Invalid API key, expired or misconfigured Microsoft Entra token, wrong token scope or audience, missing RBAC role assignment	Verify your API key or token. For Microsoft Entra ID authentication, use the https://ai.azure.com/.default scope for Azure OpenAI endpoints. See Configure keyless authentication with Microsoft Entra ID.
403	Forbidden	See HTTP 403 causes below	Depends on the specific cause
404	Not Found	Incorrect endpoint URL, deployment name doesn’t exist, unsupported API version	Verify your endpoint URL, deployment name, and API version in the REST API reference.
429	Too Many Requests	Rate limit exceeded, system capacity throttling, or a temporary rate limit adjustment	Follow the retry-after-ms header when present. Review quotas and limits and 429 throttling guidance.
500 / 503	Server Error	Transient service issue, backend overload	Retry with exponential backoff. If errors persist for more than a few minutes, check Azure Service Health or file a support request.

HTTP 403 Forbidden

A 403 response means the server understood your request but refuses to authorize it. Unlike 401, a 403 indicates that your credentials might be valid but access is denied.

Network access denied

The resource’s network access controls blocked your request.

Symptoms:

• Error message references network access, IP restriction, or private endpoint.
• You recently changed virtual network, firewall, or private endpoint settings.

Resolution:

1. In the Azure portal, go to your resource and select Networking.
2. Verify your client IP is in the allowed list if you use IP firewall rules.
3. If you use a private endpoint, verify DNS resolves to the private IP. See Troubleshoot private endpoint connectivity.
4. Check that network security group (NSG) rules on the subnet aren’t blocking traffic.

To set up a Foundry private endpoint, see Configure private link. For virtual network, firewall, and IP rules, see Configure Azure AI services virtual networks.

Resource suspended (abuse policy)

Automated abuse detection systems suspended your resource because it violated the Microsoft Online Services Acceptable Use Policy.

Symptoms:

• Error message contains “resource has been temporarily blocked”.
• All requests to the resource return 403.

Resolution: Create a support request and include:

• Your subscription ID and resource name.
• A description of your use case.
• Corrective actions you already took.

For more information, see Abuse monitoring.

Related content

• Azure OpenAI REST API reference
• Quotas and limits
• 429 throttling guidance
• Content filtering
• Abuse monitoring
• Azure Service Health