This is a concise, actionable anti-ban checklist for Claude users — especially those operating from regions where Anthropic does not officially sell Claude. It distills the most common failure modes from community reports and public reverse-engineering into five areas you can verify today. For deep dives, follow the linked guides in each section.
1. Account & Payment Safety
- One account, one identity stack: Do not reuse the same virtual-card BIN, phone number, or billing address across multiple Pro accounts within 24 hours. Since Anthropic's Feb 2026 distillation report and the Aug 31, 2026 thinking-block replay restrictions, linked disables via payment, IP, and distillation fingerprinting are more aggressively enforced.
- Corporate ownership matters: Since Sept 2025, entities more than 50% owned by companies headquartered in unsupported regions (e.g. China) are barred — even if incorporated in Singapore or the US.
- Align geography: Registration IP, billing country, and card BIN should tell a consistent story. A US card with a Singapore IP and a China timezone is a red flag.
- Avoid VoIP numbers: Use physical SIM or reputable SMS services; many VoIP prefixes fail verification or trigger instant review.
- Let subscriptions settle: Do not chargeback immediately on failed renewals — that often triggers permanent disable rather than a reversible refund.
→ Full guide: Safe Account Registration & Payment
2. Environment & Browser Hygiene
- Fix timezone first: System timezone is the heaviest single signal Claude Code reads. Move off Asia/Shanghai and confirm
getTimezoneOffset()is not UTC+8. - Language & fonts: Remove zh-CN from the top of
navigator.languages; isolate Chinese vendor fonts (MiSans, HarmonyOS Sans) in a clean browser profile. - Stop WebRTC leaks: Disable or restrict WebRTC so your real IP does not bypass the proxy.
- Use residential IP: Datacenter IPs score poorly on IP reputation checks. Prefer clean residential or mobile IPs in a supported region.
→ Full guide: Environment Cleanup & IP Setup · Browser Configuration
3. API & Claude Code Usage
- Match CLI and browser: Export the same
TZ, proxy, andANTHROPIC_BASE_URLin every shell where Claude Code runs. Split-brain configs (browser on VPN, CLI direct) are a top cause of 403 errors. - Respect rate limits: API tiers are now Start / Build / Scale (formerly Tier 1–4). Start allows ~1,000 RPM per model — but high-repetition or coordinated traffic patterns can still trigger review regardless of quota headroom. Use exponential backoff, Batch API, or prompt caching for bulk work.
- Choose relays carefully: Gateways that strip
cache_controlmarkers oranthropic-betaheaders, or inject text, break prompt cache and can look like abuse. - No credential sharing: Never paste API keys into public repos, chat logs, or shared CI logs. Rotate keys if exposed.
→ Full guide: Claude Code & API Safety · Automation Safety
4. Usage Patterns & Frequency
- Warm up new accounts: Avoid hundreds of API calls or marathon Claude Code sessions on day one. Gradual usage looks more natural than instant max-tier load.
- Avoid distillation-like patterns: Narrow, high-volume capability targeting (e.g. thousands of near-identical coding prompts) is now actively flagged. Accounts created after Aug 31, 2026 face stricter thinking-block replay binding — never cache and replay raw thinking blocks. Spread workloads across time and vary prompt structure.
- Separate workloads: Do not run aggressive scraping, mass account creation, or policy-edge prompts on your primary paid account.
- One device fingerprint per account: Switching between a China-locale laptop and a US VPS on the same login within hours raises association risk.
- Log and audit: Keep request volume and error rates visible so 429 spikes are caught before they escalate to review.
→ Full guide: Multi-Account Management · Ban Case Studies
5. When Something Goes Wrong
- Diagnose the ban type: 403 at login (IP/geo), forced refund email (payment mismatch), or full disable (policy) — each needs a different response.
- Appeal via official channels: Disabled accounts → claude.ai/restricted; API warnings → usersafety@anthropic.com. H1 2026 data shows ~10.5% appeal success — include verifiable academic/dev-use proof in English.
- Do not panic-create accounts: Community reports show successful reinstatements within 1–3 days when appeals are submitted properly. Spinning up replacement accounts while pending raises association risk.
- Fail over gracefully: Keep a domestic model or local Ollama/One-API route ready so production does not halt during recovery.
→ Full guide: Account Appeal & Recovery SOP · Domestic Alternatives
Quick Self-Check
Run the Fuck Claude detector on the same browser profile you use for Claude. Score above 60? Fix timezone and language before touching payment or API settings. Score below 30 but still banned? The issue is almost certainly IP reputation, payment BIN, or usage pattern — not browser fonts alone.