🌐 เอกสารภาษาไทยกำลังจัดทำ — เนื้อหาด้านล่างเป็นภาษาอังกฤษชั่วคราว จนกว่าจะมีการแปล. This page is not yet translated; English content is shown temporarily.

Blocked requests

If a request is rejected by a guardrail (not a budget), it returns 403 and appears on your Blocked requests page, where you can see why and report a false positive.

Prerequisite

You're signed in (Get access).

Why a request is blocked

Guardrails screen each request before it reaches a model. A request is blocked when it matches:

• a PII rule (sensitive data like an email, phone number, or key), or
• a prompt-injection rule (pattern-based or semantic) intended to subvert the model.

A blocked request never reaches the provider, so it costs nothing against your budget.

Review and report

1. Open Blocked requests in the console.
2. Find the request — you'll see the time and the rule that matched.
3. If it was wrongly blocked, click Report to flag it as a false positive. Your org admin can then tune the guardrail.

Distinguish 403 from 429

A 403 means a guardrail blocked the content. A 429 means you hit a budget or token limit — see Usage & budget. The request lifecycle table maps each status to its gate.

Next steps

• Administrators tune rules in Guardrails.