AWS bedrock: Claude Messages API: expand stop_details.category values and partial-output guidance
Summary
Enumerates the full set of stop_details.category values (cyber, bio, frontier_llm, reasoning_extraction, general_harms, null), notes the category determines billing, and changes mid-stream refusal guidance to treat streamed content as incomplete and discard it.
Security assessment
The edit documents API response fields and billing-related refusal handling; the 'discard partial output' wording is correctness guidance for consumers, not a security control or vulnerability fix.
Evidence
+ * **category** (string or null) – The content category that triggered the restriction: `"cyber"`, `"bio"`, `"frontier_llm"`, `"reasoning_extraction"`, `"general_harms"`, or `null`. `null` is a valid final value, not a placeholder. The category determines whether the refused request is billed. For details, see [Billing notes](./model-cards-anthropic.html#model-cards-anthropic-billing-notes).
Diff
diff --git a/bedrock/latest/userguide/model-parameters-anthropic-claude-messages-request-response.md b/bedrock/latest/userguide/model-parameters-anthropic-claude-messages-request-response.md index 3aa442a..109ecfc 100644 --- a//bedrock/latest/userguide/model-parameters-anthropic-claude-messages-request-response.md +++ b//bedrock/latest/userguide/model-parameters-anthropic-claude-messages-request-response.md @@ -351 +351 @@ Example responses with new stop_reason values: - * **category** (string or null) – The content category that triggered the restriction (e.g., `"cyber"`, `"bio"`). Null when the refusal does not map to a named category. + * **category** (string or null) – The content category that triggered the restriction: `"cyber"`, `"bio"`, `"frontier_llm"`, `"reasoning_extraction"`, `"general_harms"`, or `null`. `null` is a valid final value, not a placeholder. The category determines whether the refused request is billed. For details, see [Billing notes](./model-cards-anthropic.html#model-cards-anthropic-billing-notes). @@ -357 +357 @@ Example responses with new stop_reason values: -Branch on `stop_reason`, not on `stop_details`. The `stop_details` field is informational and can be null even when `stop_reason` is `"refusal"`. Refusals can also occur mid-stream: the classifier may fire after some content has already been streamed. In streaming mode, the refusal arrives as the final `message_delta` event with `stop_reason: "refusal"`; any content blocks streamed before the refusal are valid partial output. +Branch on `stop_reason`, not on `stop_details`. The `stop_details` field is informational and can be null even when `stop_reason` is `"refusal"`. Refusals can also occur mid-stream: the classifier may fire after some content has already been streamed. In streaming mode, the refusal arrives as the final `message_delta` event with `stop_reason: "refusal"`; treat any content streamed before the refusal as incomplete and discard it.