Source: Unite.AI
Amazon Web Services published guidance on its Machine Learning Blog on October 5, 2026, detailing how organizations with regulatory or compliance requirements, including the International Traffic in Arms Regulations (ITAR), can run Anthropic’s Claude Code with Claude Opus 5.5 and Claude Sonnet 5.5 on Amazon Bedrock in the AWS GovCloud (US) Regions.
According to the post, Claude Opus 5.5 and Claude Sonnet 5.5 hold FedRAMP Class D (formerly High) certification on Amazon Bedrock, while Claude Sonnet 5 holds FedRAMP Class D certification and DoD Impact Level 4 and 5 (IL4/IL5) authorization. Bedrock in GovCloud (US) offers FedRAMP Class D and DoD Cloud Service Provider SRG IL4/IL5 authorization pathways, and its built-in data protection means customer content is not stored, logged, used to train AWS models, or shared with third parties.
Claude Opus 5.5 became available on AWS GovCloud (US) on September 22, 2026, according to an AWS announcement, and Claude Sonnet 5.5 followed on September 28, 2026. In the Opus announcement, AWS said the model is offered on Amazon Bedrock with zero data retention support by default and regional data residency.
Two Bedrock Endpoint Surfaces
Amazon Bedrock in AWS GovCloud (US) supports two endpoint surfaces, bedrock-runtime and bedrock-mantle, both powered by the same Mantle inference engine with a Zero Operator Access architecture. The bedrock-runtime endpoint uses the AWS SDK through the InvokeModel and Converse APIs and supports Amazon Bedrock Guardrails, Knowledge Bases, Agents, and invocation logging; AWS recommends it for most new applications, particularly those requiring audit trails.
The bedrock-mantle endpoint supports the Anthropic Messages API natively, with capabilities exclusive to that surface, such as server-side tools, background inference, and Projects. The bedrock-runtime endpoint is available in both GovCloud Regions (US-West and US-East), while bedrock-mantle is available only in AWS GovCloud (US-West). Guardrails and invocation logging are exclusive to bedrock-runtime, so the post directs deployments requiring comprehensive audit trails and content filtering to that endpoint.
What Claude Code Does
Claude Code is Anthropic’s agentic coding tool that reads a codebase, edits files, runs commands, and integrates with development tools. It operates in the terminal, in integrated development environments such as VS Code and JetBrains, and in the background through the Claude Agent SDK.
AWS lists capabilities that include writing code and fixing bugs across multiple files, executing and fixing tests and linting, searching Git history, resolving merge conflicts, and creating commits and pull requests. The tool connects to external tools and data sources through the Model Context Protocol, including the AWS Command Line Interface, Terraform, and Kubernetes, and can spawn sub-agents that work on different parts of a task simultaneously. Behavior can be customized with CLAUDE.md memory files, skills, and hooks, and recurring tasks can be automated through CI/CD integration with GitHub Actions or GitLab CI/CD.
Setup and Configuration
Prerequisites include an AWS GovCloud (US) account with Amazon Bedrock access; IAM permissions that at minimum include bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, bedrock:ListInferenceProfiles, and bedrock:GetInferenceProfile for bedrock-runtime, plus bedrock-mantle actions or the AmazonBedrockMantleInferenceAccess managed policy for the Mantle endpoint; model access to Claude Opus 5.5, Claude Sonnet 5.5, and Claude Sonnet 5 enabled in the account; and the AWS CLI configured with short-term credentials or AWS SSO.
The post documents three configuration paths, labeled A through C. Option A uses Claude Code’s interactive login wizard: select 3rd-party platform, then Amazon Bedrock, choose the us-gov-west-1 region, and pin models; previously configured installations can reopen the wizard with /setup-bedrock. Option B sets environment variables: CLAUDECODEUSEBEDROCK=1, AWSREGION set to us-gov-west-1, and ANTHROPICMODEL pointing to us-gov.-prefixed model IDs such as us-gov.anthropic.claude-sonnet-5-5 or us-gov.anthropic.claude-opus-5-5. Option C routes Claude Code through Mantle with CLAUDECODEUSEMANTLE=1, and the Bedrock and Mantle flags can be combined in a single session. Running /status verifies the setup; the provider line shows Amazon Bedrock or Amazon Bedrock (Mantle).
Enterprise Deployment and Cost Controls
With Claude Code now generally available, AWS recommends governing identity and access through AWS IAM Identity Center with temporary, role-based credentials rather than static access keys, automating default environment variables or managing settings files centrally, and implementing its Guidance for Claude Code with Amazon Bedrock for large enterprise deployments. AWS also advises reviewing Bedrock service quotas for tokens per minute and requests per minute against the number of active developers.
AWS recommends pinning model versions because unpinned aliases such as sonnet and opus resolve to Claude Code’s built-in defaults, which may change between releases. Claude Code defaults to Claude Opus 5.5 as its primary model, so an unpinned deployment is billed at the Opus per-token rate, which the post notes is higher than Claude Sonnet 5.5’s. For cost governance, AWS points to per-user token guardrails enforcing daily limits with alerts at 80% and 100% thresholds, built on CloudWatch invocation logging, Lambda, and DynamoDB, and to prompt caching with 5-minute and 1-hour TTL options for supported models.
AWS suggests defaulting teams to Claude Sonnet 5.5 and reserving Claude Opus 5.5 for tasks requiring deeper reasoning or longer autonomous runs, while workloads requiring IL4/IL5 authorization should default to Claude Sonnet 5, which the post describes as offering the strongest compliance coverage for sensitive environments. It characterizes Claude Sonnet 5.5 as a step up from Claude Sonnet 5 on coding and knowledge work at a lower cost per task. According to Anthropic, as cited in the AWS records, Claude Opus 5.5 completes tasks using fewer tokens than Claude Opus 5 at a lower price per token.
On security, AWS recommends conducting a thorough assessment before deployment: Amazon Bedrock secures the inference layer, but Claude Code runs on local developer machines and requires separate evaluation and risk management. The post advises applying controls such as managed permissions, invocation logging, and per-user token limits, as it would for other third-party software in its environment.