Anthropic Supported Regions Policy: Recover Work After a Suspension

Understand the November 12, 2026 Supported Regions Policy, distinguish verification and billing failures from suspensions, preserve local Claude Code work, and continue with an eligible non-Anthropic tool and model.

Contents
Anthropic Supported Regions Policy: Recover Work After a Suspension

If Claude suspends your account citing “Supported Countries Policy,” preserve your local work before changing providers. Anthropic’s Supported Regions Policy covers individual developers and teams using Claude Code, APIs, cloud platforms, and authorized resellers. Check your location and organization’s eligibility, appeal an apparent mistake, and prepare a handoff to an eligible non-Anthropic tool and model.

Check the three eligibility rules

As of 2026-11-19, the Supported Regions Policy is marked effective November 12, 2026. It covers Claude.ai, Claude Code, Cowork, the API and developer platform, and Claude models supplied through cloud platforms and authorized resellers.

The policy identifies three unsupported situations:

  1. A person uses the service while physically in an unsupported region. This includes personnel of otherwise supported organizations.
  2. An organization is incorporated or headquartered in an unsupported region. Its users and personnel remain covered even when physically elsewhere.
  3. An organization is majority-owned or controlled, directly or indirectly, by persons or entities in unsupported regions. The same condition follows its users and personnel.

A supported-country office does not resolve an unsupported headquarters or ownership/control relationship. The wording is “majority-owned or controlled”, so do not reduce the check to shareholding alone. Ask your legal or compliance team to review unclear control relationships.

The current supported-region list does not include Russia, Belarus, Iran, mainland China, Hong Kong, or Macau. Taiwan is listed separately. Ukraine is listed except for Crimea, Donetsk, Kherson, Luhansk, and Zaporizhzhia regions.

Anthropic’s October 8 announcement describes the regional update as a clarification of enforcement. Do not treat November 12 as evidence that every restriction began that day.

Changing a Claude API endpoint, cloud account, or reseller does not create an exception to these rules. A different interface that still uses Claude is not a non-Anthropic alternative; see the account dependencies in OpenCode’s Claude Agent integration.

Separate verification, payment, and network symptoms

A suspension notice is more informative than the last action before it. Keep its exact wording, timestamp, affected account, and whether the restriction affects one organization or your whole account.

Anthropic’s identity-verification guidance describes verification as a policy and compliance control. Passing an identity check does not itself establish that every regional eligibility condition is satisfied.

In a November 14, 2026 post attributed to @dimarus, Dmitry Mashkin reported suspension after Russian-passport verification and a rejected appeal. This is one user’s account, not proof of a passport-only rule or a measured ban rate. The published regional clauses refer to location and entity relationships, not nationality alone.

For earlier context, our guide to Anthropic’s June 2026 identity-verification notices distinguishes those notices from a universal passport requirement.

Payment and network problems need separate diagnosis. Anthropic’s card-decline guidance includes billing-country consistency checks. Its login-error guidance discusses VPN troubleshooting for generic login failures. Neither establishes that a particular payment change or IP address caused a regional-policy suspension.

Record recent travel, network changes, and payment events for support, but do not label them confirmed triggers. Buying another plan, changing identity details, or hiding your location is not a supported remedy for ineligibility.

Appeal a mistake and request account data

Use Anthropic’s current appeal instructions, which direct suspended users to sign in and use the appeal form. Follow the options actually presented on the restricted-account screen.

Include your account identifier, the exact notice, actual location, and relevant incorporation, headquarters, or ownership facts. Ask which condition applies and request review of any factual error. Submit identity documents only through an official requested verification flow, not a public thread.

Use any export option offered on that screen before deleting the account. In an accessible personal-account interface, Settings > Privacy > Export data starts an export delivered through an email download link. Do not assume an account export replaces your repository backup.

Plan the migration in parallel rather than assuming a reinstatement date. Handle billing disputes and identity-data requests separately from the technical recovery.

Preserve Claude Code work without another model request

Save the working files and local state

Stop tasks that are still writing files, then make a private backup of the complete working directory. Include hidden files, untracked files, and each active worktree, not only the last Git commit. For a linked worktree, preserve the main repository too.

Claude Code’s session documentation places local transcripts at ~/.claude/projects/<project>/<session-id>.jsonl by default. Copy the entire relevant project directory, rather than only one transcript. Also preserve the rest of your Claude configuration directory privately before uninstalling or clearing anything.

If you used CLAUDE_CONFIG_DIR, inspect that directory instead. On Windows, the default ~/.claude location means %USERPROFILE%\.claude, as the settings reference explains. Perform this backup on the machine or remote environment where Claude Code actually ran; a local copy cannot capture server-only files.

Keep CLAUDE.md, the project’s .claude/ directory, and .mcp.json when present. Review .claude/settings.json, .claude/settings.local.json, and ~/.claude/settings.json before translating permissions or hooks. The MCP configuration reference places project MCP definitions in .mcp.json; also inspect the default global file, ~/.claude.json, for other integrations.

The local-storage reference warns that transcripts are not encrypted at rest. Keep backups private and redact secrets before sharing any excerpt. Do not upload an entire configuration directory or import old authentication credentials into the replacement tool.

Make a portable task handoff

If your existing Claude Code interface still permits local commands, use the documented conversation export command:

/export recovery-session.txt

Open the resulting file and confirm that it contains the relevant conversation. If sign-in blocks the interface or export fails, keep the JSONL backup and inspect a copy in a text editor. Do not depend on a new model-generated summary to rescue the task.

Create HANDOFF.md in the project yourself. Record the unfinished goal, acceptance criteria, changed files, actual test commands and results, unresolved failures, and the next small action. Add approved project instructions from CLAUDE.md; describe required integrations without copying secrets.

Verify this note against the current files. A prior assistant message saying “tests passed” is not a substitute for the test output. You are transferring task context and working files, not promising to restore running subagents in a different client.

Choose an eligible route before paying again

Choose a route based on the actual blocker. Confirm the new provider’s regional and business eligibility before creating an account or uploading code.

RouteRequirements and billingChoose it when
Appeal the existing Claude restrictionYour existing account and evidence that the relevant person and organization satisfy the policy. An appeal does not change regional eligibility.The restriction appears factually wrong and preserving Claude access matters.
A non-Anthropic provider’s own API, such as DeepSeekYour own provider account and API key, an eligible location/entity, and an accepted payment method. DeepSeek meters model usage by input/output tokens.Direct vendor billing meets your requirements.
BetterToken with a non-Anthropic modelYour own BetterToken account, API key, and accepted top-up method; eligibility still needs checking. Billing follows the model’s current input, output, and cache rules.You want pay-as-you-go model access through its documented Aider integration.
A local model through Aider and OllamaSuitable hardware and a model you may download and use under its license. Local inference has hardware costs rather than a hosted model token bill.A hosted account is unsuitable and the local model handles your task adequately.

DeepSeek documents its API access and token billing. Aider documents its local Ollama connection. These are different choices from routing Claude through another provider.

Example: Aider with BetterToken and deepseek-flash

As of 2026-11-19, BetterToken’s model catalog lists deepseek-flash for its OpenAI-compatible API. Its pay-as-you-go API route uses your own account and key. BetterToken is not an Anthropic product, and API access is not a Claude subscription.

Create a key in the workspace and check current input, output, and cache prices before funding usage. Available payment channels and fees depend on the console at payment time.

Follow the BetterToken Aider configuration with Base URL https://www.bettertoken.ai/v1. With Python installed, use the Aider installer. From the project directory containing your reviewed HANDOFF.md, run this in Bash:

python -m pip install aider-install
aider-install

read -rs BETTERTOKEN_API_KEY && export BETTERTOKEN_API_KEY
export OPENAI_API_KEY="$BETTERTOKEN_API_KEY"
export OPENAI_API_BASE="https://www.bettertoken.ai/v1"

aider --model openai/deepseek-flash --chat-mode ask --read HANDOFF.md

The openai/ prefix selects Aider’s compatible-API route; it does not mean the model is from OpenAI. The read command waits for your key without displaying it. Paste the key and press Enter; it stays in the shell environment rather than the command text.

Aider’s ask mode discusses code without editing it, and --read supplies the handoff as read-only context.

Create a BetterToken account

Verify the handoff before allowing edits

In Aider, submit:

Read HANDOFF.md. Restate the unfinished goal, identify the changed files, and propose the next test. Do not edit source files.

Check that the selected model is openai/deepseek-flash and that its answer matches your saved work. For an authentication failure, follow BetterToken’s Aider troubleshooting guidance to check the key and endpoint before retrying. For a model-availability error, check the current catalog ID.

If Aider reports an unknown context window or token cost, consult its model-warning guidance. Do not treat an unknown-model cost estimate as the provider’s tariff.

After the handoff is accurate, use /add for the specific files needed and /chat-mode code to permit editing. Request one small change, run the project’s tests yourself, and compare the source diff against your saved baseline. Retain the recovery copy until the resumed task passes its acceptance criteria.

Ready to optimize your LLM workflow?

Join thousands of developers building faster, smarter, and more cost-effective AI applications with BetterToken.

Get Started for Free