AWS bedrock high security documentation change
Summary
Updated data retention policy details: replaced 'prompts and outputs' with 'inputs and outputs', added retention requirements for Claude Fable 5 including mandatory opt-in for sharing with Anthropic, clarified storage locations and third-party sharing policies.
Security assessment
The changes explicitly address data handling security policies including mandatory retention periods, third-party data sharing requirements, and cross-region storage implications. The addition of Claude Fable 5's requirement for customer opt-in to share data with Anthropic for human review demonstrates concrete security controls.
Diff
diff --git a/bedrock/latest/userguide/abuse-detection.md b/bedrock/latest/userguide/abuse-detection.md index a483856e2..a8d1ae08f 100644 --- a//bedrock/latest/userguide/abuse-detection.md +++ b//bedrock/latest/userguide/abuse-detection.md @@ -13 +13 @@ Amazon Bedrock uses a zero operator access (ZOA) data security model. This means -However, for specific abuse detection purposes related to the following models, we may be required to store prompts and outputs: +However, for specific abuse detection purposes related to the following models, we may be required to store inputs and outputs: @@ -15 +15 @@ However, for specific abuse detection purposes related to the following models, - * For GPT-5.4 and GPT-5.5, classifier-flagged traffic will be retained for up to 30 days for automated offline abuse detection. For these models, eligible customers may request full ZDR through their AWS account team. + * For OpenAI GPT-5.4 and GPT-5.5, classifier-flagged traffic will be retained for up to 30 days for automated offline abuse detection. @@ -16,0 +17 @@ However, for specific abuse detection purposes related to the following models, + * For Anthropic Claude Fable 5, inputs and outputs will be retained for up to 30 days. In order to use Claude Fable 5, as required by Anthropic, you must opt in to sharing retained traffic with Anthropic for abuse detection and potential human review. @@ -19,0 +21,3 @@ However, for specific abuse detection purposes related to the following models, + +Retained inputs and outputs are stored and processed by AWS and are not shared with third-party model providers, unless you opt in to sharing with the model provider. If cross-region inference is enabled for these models, retained inputs and outputs are stored in destination regions. For these models, eligible customers may request full ZDR through their AWS account team. +