Claude Code now runs on Amazon Bedrock in AWS GovCloud (US) for regulated work

Models and compliance status

AWS says the availability of Anthropic Claude Opus 5.5 and Claude Sonnet 5.5 in the AWS GovCloud (US) Regions opens the door to AI-assisted development for workloads with regulatory or compliance requirements, including International Traffic in Arms Regulations (ITAR).

On certifications, the post states that Claude Sonnet 5 holds FedRAMP Class D (formerly High) certification and DoD Impact Level 4 and 5 (IL4/IL5) authorization. Claude Opus 5.5 and Claude Sonnet 5.5 hold FedRAMP Class D certification on Amazon Bedrock. AWS points readers to verify current model certification status and to the Amazon Bedrock models compliance page. It also cautions that the approach may not suit every organization or compliance program, and that readers should evaluate it against their own obligations.

Two endpoints on Amazon Bedrock

Amazon Bedrock in AWS GovCloud (US) offers two endpoint surfaces, bedrock-runtime and bedrock-mantle. Both run on the same Mantle inference engine with a Zero Operator Access (ZOA) architecture, according to AWS.

• bedrock-runtime uses the AWS SDK (InvokeModel and Converse APIs) and supports Amazon Bedrock Guardrails, Knowledge Bases, Agents and invocation logging. AWS recommends it for most new applications, particularly those that need audit trails. It is available in both GovCloud Regions, US-West and US-East.
• bedrock-mantle supports the Anthropic Messages API natively and offers capabilities currently available only on that surface, such as server-side tools, background inference and Projects. It is available in AWS GovCloud (US-West).

Both endpoints cover Claude Opus 5.5, Claude Sonnet 5.5 and Claude Sonnet 5. AWS also notes that Guardrails and invocation logging are available exclusively through bedrock-runtime, so deployments that need those compliance features should use it.

What Claude Code does and how to set it up

Claude Code is Anthropic's agentic coding tool. It reads a codebase, edits files, runs commands and connects to development tools. It works in the terminal, in IDEs such as VS Code and JetBrains, and in the background through the Claude Agent SDK. AWS lists tasks including fixing bugs across multiple files, answering questions about architecture, running tests and linting, resolving merge conflicts, creating commits and pull requests, and spawning sub-agents. It can connect to tools such as the AWS CLI, Terraform and Kubernetes through the Model Context Protocol, and integrate with CI/CD through GitHub Actions or GitLab CI/CD.

Prerequisites include an AWS GovCloud (US) account with Bedrock access, suitable IAM roles and permissions, model access enabled for the three Claude models, and an AWS CLI configured with valid session credentials. Setup offers three routes:

• Option A: an interactive wizard, recommended by AWS, started with the /login command. Users pick Amazon Bedrock under 3rd-party platform, choose an authentication method, set the region to us-gov-west-1 and pin models.
• Option B: manual environment variables for scripted or enterprise deployments, such as CLAUDE_CODE_USE_BEDROCK=1, AWS_REGION and ANTHROPIC_MODEL set to us-gov.anthropic.claude-sonnet-5-5 or us-gov.anthropic.claude-opus-5-5.
• Option C: routing through the Mantle endpoint with CLAUDE_CODE_USE_MANTLE=1.

Users can confirm the setup with the /status command, where the provider line should show Amazon Bedrock or Amazon Bedrock (Mantle).

What it means for regulated teams

For organizations bound by frameworks such as FedRAMP, IL4/IL5 or ITAR, the post sets out a path to use an agentic coding assistant inside GovCloud. AWS says Bedrock does not store, log or use customer content to train AWS models, and that it integrates with existing security controls in GovCloud.

For wider rollouts, AWS suggests using AWS IAM Identity Center to centrally govern access to Claude Code, giving developers temporary, role-based credentials instead of static access keys. The source text is cut off before the remaining deployment recommendations. Because certification status varies by model, teams should check the current model-level status before relying on any particular model.