Claude Code Auto Mode on AWS Bedrock, Google Vertex AI & Azure Foundry

Enterprise teams running Claude Code on AWS, Google Cloud, or Azure just got one of the most requested upgrades of 2026: Auto Mode is now available on Bedrock, Vertex AI, and Azure Foundry for Opus 4.7 and Opus 4.8. One environment variable unlocks the same permission-free agentic loop your developers have been using on the Anthropic API — routed through infrastructure your security team has already approved.

If you've been waiting to roll out Claude Code at scale because of data residency concerns, cloud spend consolidation, or enterprise compliance requirements, this changes the calculus entirely.

What Is Claude Code Auto Mode?

Before diving into the enterprise setup, a quick recap of what Auto Mode actually does.

By default, Claude Code asks for permission before running each tool — reading files, executing commands, editing code. That's safe for solo developers learning the tool. It becomes friction at production scale, when you want Claude Code to run a full test-fix-retest loop unattended.

The result: you describe the task, Claude Code executes the full loop, and you come back to a completed diff. That's the same experience developers on Anthropic's Max, Team, and Enterprise plans have had since early 2026. As of May 30, 2026 (Claude Code v2.1.15x), it's available on the three major enterprise cloud platforms.

What's New: Auto Mode Lands on Bedrock, Vertex, and Foundry

Claude Code v2.1.15x shipped with a single configuration toggle that unlocks Auto Mode across Bedrock, Vertex AI, and Azure Foundry:

bashCLAUDE_CODE_ENABLE_AUTO_MODE=1

Set that environment variable, point Claude Code at your existing Bedrock or Vertex integration using Opus 4.7 or Opus 4.8, and the agentic loop runs inside your cloud perimeter — no Anthropic API calls leaving your environment.

This matters for several distinct enterprise reasons:

How to Enable Claude Code Auto Mode on Enterprise Platforms

AWS Bedrock

You'll need an active Bedrock model access grant for anthropic.claude-opus-4-7 or anthropic.claude-opus-4-8 in your target region.

bash# Configure your AWS credentials first
export AWS_REGION=us-east-1
export AWS_PROFILE=your-bedrock-profile

# Enable Auto Mode
export CLAUDE_CODE_ENABLE_AUTO_MODE=1

# Point Claude Code at Bedrock
export ANTHROPIC_API_KEY=""  # leave empty; Bedrock uses AWS credentials
export CLAUDE_CODE_USE_BEDROCK=1
export ANTHROPIC_MODEL=anthropic.claude-opus-4-8-20251001-v1:0

claude "Refactor the authentication module to use JWT refresh tokens"

Bedrock logs flow to CloudTrail (API calls) and CloudWatch Logs (model invocations if you've enabled model invocation logging in Bedrock settings). This gives you a complete audit trail for every Auto Mode turn — which prompts were sent, which tools were called, token counts.

Google Cloud Vertex AI

You'll need Vertex AI API enabled and the Claude model enabled in Model Garden for your project.

bash# Google Cloud credentials
export GOOGLE_CLOUD_PROJECT=your-project-id
export GOOGLE_CLOUD_REGION=us-central1
gcloud auth application-default login

# Enable Auto Mode
export CLAUDE_CODE_ENABLE_AUTO_MODE=1

# Point Claude Code at Vertex
export CLAUDE_CODE_USE_VERTEX=1
export ANTHROPIC_VERTEX_PROJECT_ID=your-project-id
export ANTHROPIC_VERTEX_REGION=us-central1
export ANTHROPIC_MODEL=claude-opus-4-8@20251001

claude "Add OpenTelemetry instrumentation to the API gateway"

Vertex logs flow to Cloud Logging and optionally Cloud Audit Logs. If you're using VPC Service Controls, Vertex AI is a supported service — Claude Code Auto Mode sessions will respect your perimeter rules.

Azure Foundry (Azure AI Services)

Azure Foundry requires a Claude deployment in your Azure AI hub.

bash# Azure credentials
az login
export AZURE_SUBSCRIPTION_ID=your-sub-id

# Enable Auto Mode
export CLAUDE_CODE_ENABLE_AUTO_MODE=1

# Point Claude Code at Foundry
export CLAUDE_CODE_USE_FOUNDRY=1
export AZURE_CLAUDE_ENDPOINT=https://your-resource.services.ai.azure.com/
export ANTHROPIC_MODEL=claude-opus-4-8

claude "Migrate the data pipeline from pandas to Polars"

Azure logs flow to Azure Monitor and Log Analytics. You can build Azure Monitor dashboards to track Auto Mode usage, latency, and token consumption per workspace.

Observability: One Flag, Three Audit Trails

The CLAUDE_CODE_ENABLE_AUTO_MODE=1 flag is all-or-nothing — there's no per-provider toggle. But observability splits across platforms, and that's actually an advantage if you're already invested in one cloud's monitoring stack.

Each platform exposes different granularity. Bedrock's model invocation logs capture input/output token counts per turn, which is useful for chargebacks if you're running Auto Mode across multiple teams. Vertex's Cloud Logging can capture full request/response payloads (enable carefully — this logs your prompts and code). Azure Monitor has the richest alerting ecosystem if your team already uses Azure Sentinel.

Combining Auto Mode with Claude Code's New Production Features

Auto Mode on enterprise platforms pairs well with two other features shipped in June 2026:

json{
  "model": "anthropic.claude-opus-4-8-20251001-v1:0",
  "fallbackModel": ["anthropic.claude-sonnet-4-6-20251001-v1:0"]
}

bashCLAUDE_CODE_ENABLE_AUTO_MODE=1 claude --safe-mode "Reproduce the issue"

Safe mode disables CLAUDE.md, skills, plugins, hooks, and MCP servers for that session. If the problem disappears, you've found your culprit.

Enterprise Readiness: What Auto Mode Can and Can't Do Today

Auto Mode on Bedrock/Vertex/Foundry is production-ready for structured pilots — single teams, defined repositories, bounded scope. It's not ready for unrestricted broad rollout across a large engineering org without additional guardrails.

• Automated test-fix loops on isolated service repos
• Code migration tasks (upgrading libraries, changing APIs) with a human reviewing the final diff
• Documentation generation where the worst-case outcome is a bad doc, not a bad deploy
• Pre-commit hooks that trigger Claude Code to check and fix failing tests before merge

• Auto Mode in CI/CD with write access to main — gate behind a human approval step
• Cross-repository refactors where a single session might touch 50+ files
• Auto Mode on production infrastructure repositories without read-only mode configured

The pattern that works: give Claude Code Auto Mode a feature branch, not main. Let it run the full loop on the branch, then require a human PR review before merge. You get the speed of autonomous operation without the risk of direct production changes.

Why This Matters for the Claude Ecosystem

The Bedrock/Vertex/Foundry expansion removes the last major enterprise objection to deploying Claude Code at scale. The previous blockers were:

What remains is cultural adoption: getting engineering teams comfortable with AI making multiple sequential decisions unattended. That's a training and governance challenge, not a technical one — and it's solvable.

If you're preparing for the Claude Certified Architect (CCA) certification, understanding how to deploy Claude Code in enterprise environments — including platform choice, observability configuration, and governance patterns — is directly relevant to the architecture domains covered in the exam.

Key Takeaways

• Claude Code Auto Mode is now available on AWS Bedrock, Google Vertex AI, and Azure Foundry for Opus 4.7 and Opus 4.8
• Enable with a single environment variable: CLAUDE_CODE_ENABLE_AUTO_MODE=1
• Data stays inside your existing cloud perimeter — no Anthropic API calls required
• Consumption counts toward your existing AWS/GCP/Azure cloud spend commitments
• Pair with fallback model chains and safe mode for production reliability
• Start with feature-branch-scoped pilots; don't give Auto Mode direct write access to main without a review gate

Next Steps

If you're an enterprise team evaluating Claude Code deployment, the fastest path to a production pilot is:

And if you want to deepen your understanding of how Anthropic's enterprise deployment models work — including Bedrock vs. direct API trade-offs, agentic architectures, and platform-specific constraints — our CCA certification practice tests cover these topics as part of the Claude Certified Architect exam prep curriculum.