Configure delegated access for AWS Bedrock

Overview

For AWS Bedrock, delegated access usually means using IAM role assumption (STS) and short-lived credentials instead of distributing long-lived AWS keys.

Bedrock APIs are generally authenticated with SigV4 signing. Flashback's AI LLM configuration accepts provider endpoint + secret/token fields, so production integrations often use a controlled credential flow or a signing gateway.

When to use this

• Production workloads with strict key management requirements.

• Teams using centralized IAM and short credential lifetime.

• Multi-team environments where external access must be auditable and revocable.

Prerequisites

• AWS account with Bedrock enabled in your target region.

• Permissions to create/update IAM roles and trust policies.

• Defined list of models/actions required by your workload.

• Flashback AI LLM setup path: Configure an AI LLM.

Step-by-step (provider side)

1

Create an IAM role with least privilege

Create a role for delegated Bedrock access and attach only required actions, for example:

• bedrock:InvokeModel

• bedrock:InvokeModelWithResponseStream (if streaming is needed)

• optional model discovery permissions only if your flow requires them.

Scope permissions to required model ARNs/regions whenever possible.

2

Configure trust policy with STS AssumeRole + External ID

Use a third-party access trust policy pattern (sts:AssumeRole) and require an External ID condition.

This reduces confused-deputy risk and keeps delegated access explicit and auditable.

3

Choose an integration pattern for Flashback

Because Bedrock requests are SigV4-signed, common patterns are:

1. Signing gateway/proxy pattern (recommended for many setups) Run an internal service that:

   • assumes the IAM role,

   • signs Bedrock requests with SigV4,

   • exposes a stable endpoint that Flashback can call.

2. Direct configuration pattern (only if supported in your environment) If your Flashback environment supports direct Bedrock endpoint + secret/token flow, configure only documented fields and validate.

If direct Bedrock auth behavior is not explicitly documented in your environment, treat it as an integration pattern and validate with a non-production model first.

Configure in Flashback

Use Configure an AI LLM with the existing model:

• Select the appropriate AI LLM Type for your provider/integration.

• Set API Endpoint to your provider endpoint or signing proxy endpoint.

• Set API Secret to the credential/token expected by that endpoint.

• Use API Key only if your endpoint requires it.

Do not add undocumented fields; map provider-side delegated credentials to Flashback's existing endpoint + secret/token inputs.