Help
Why was a request blocked?
Every refusal says why and links to a full decision trace. Start with the request ID your app received.
Find the request
Every response carries X-AIGov-Request-Id, and every error message ends with it. Paste it into the search box in Requests (opens in a new tab) to see each step: what was detected, which rule matched, which models were tried and why it was refused.
Common codes
| Code | What it means | What to do |
|---|---|---|
missing_api_key · invalid_api_key · expired_api_key | No key, or the key is wrong, revoked or expired. | Check the app sends Authorization: Bearer <key> (or x-api-key). Create a new key if needed. |
data_blocked | The request contained data the profile blocks, such as card numbers or secrets. | Remove the data, or use a profile that redacts it instead. |
policy_denied | No rule allows this request, or no allowed model meets the key's frameworks. | Open the decision trace to see which rule or framework refused it. Add a provider or rule that fits. |
prompt_injection | The message looked like an attempt to override instructions. | Rephrase. If it's a false alarm, set the profile's sensitivity to Balanced or Lenient. |
guard_model_blocked · output_guard_blocked | The safety model judged the request, or the answer, unsafe. | The category is in the message. Adjust which categories block on the profile if it's a legitimate use. |
denied_topic · prompt_too_large · attachments_blocked | A blocked topic, a message bigger than allowed, or an attachment the profile doesn't accept. | Edit Content limits on the profile. |
held_for_review | Risky or high-risk AI use waits for a person. | An approver can release it under Approvals. |
budget_exceeded · plan_limit_reached | The workload's monthly budget or the plan's monthly requests are used up. | Raise the budget on the workload, buy a request pack, or upgrade. |
upstream_unavailable | Every allowed model failed or timed out. | Check Providers for outages or missing API keys. |
not_governed (501) | Endpoints whose content can't be inspected first (audio transcription, batches, file uploads). | Send text instead. |
billing_required · tenant_suspended (402) | The organisation needs billing set up, or is suspended. | Contact your administrator. |
Setup problems
| Symptom | Fix |
|---|---|
| 404 or connection errors from an OpenAI-compatible tool | The base URL must end in /v1. |
| 404 from Claude Code or an Anthropic SDK | The Anthropic base URL is the gateway address without /v1. |
| "Model not found" | Use a model your key may use: GET /v1/models lists them, as does the key's panel in API keys (opens in a new tab). |
| A different model answered | Policy redirected the request for its data. X-AIGov-Model names the model; the decision trace says why. |
| The key can't reach any chat model | Add a provider in Providers & models (opens in a new tab), or check the key's profile and policy rules. |
Test before you connect
The Playground (opens in a new tab) shows the decision for any message and profile without calling a model. Use it to check a rule or profile change before it affects real traffic.