How to Avoid Getting Banned from ChatGPT, Codex and Claude (2026)

Both OpenAI (ChatGPT and Codex) and Anthropic (Claude and Claude Code) are geo-restricted, and Hong Kong is not on either supported-country list — so from the moment you sign up, you are already in a higher-risk category than a user in the US or UK. That does not mean you will be banned, but it means you cannot afford to ignore the signals that trigger bans. This guide explains what actually gets accounts killed, what is specific to each service, and the habits that keep an account alive for years.

Why the risk is real — and why Claude is stricter

OpenAI bans are most often about login environment, shared accounts and payment: an account that pops up in different countries, is shared by several people, or was paid for with a card that later gets charged back. Anthropic has gone further. In 2026 it rolled out heavier controls — KYC checks on new accounts, much tighter Stripe card screening that has batch-rejected whole families of virtual cards, and periodic ban waves that have hit users in Hong Kong and mainland China hard. Appeal success for Claude bans is reported to be very low (community estimates around 3%), so for Claude, prevention genuinely is the only reliable strategy.

Keep your VPN on — and keep it consistent

• Use the same country for signup, payment and every login. Registering in the US and suddenly appearing in Singapore is the #1 ban signal for both companies.
• Avoid free VPNs and heavily shared nodes — their IPs are datacenter IPs that both companies flag as high risk.
• Residential IP is best. A stable home/business line, or a dedicated server you always use, looks far more natural than a shared commercial VPN exit.
• Match your timezone to your declared country — a US-registered account used at 3am China time every day adds up.

Point every signal at the same country

• VPN country = phone-number country = card country = billing-address country. Mixing a US VPN with a UK card and a Japanese number is exactly the kind of mismatch fraud checks are built to catch.
• If your phone number is a UK (+44) line, keep the account and payment in a UK-consistent setup where possible.
• Do not re-verify or log in from a completely different country than the one you registered — wait until you can use the same node.

Payment hygiene (the silent ban trigger)

• Never share an account. “拼车”/shared accounts are detected by multiple IPs and devices and get the whole group banned at once.
• Avoid 代充 sold at unrealistically low prices — those often use stolen cards; when the card is charged back, every linked account is banned.
• Watch out for virtual cards with widely-flagged card BINs. Stripe now batch-rejects whole families of these — and can retroactively claw back payments, banning accounts that were fine for months.
• Prefer the App Store / Google Play in-app route, which routes billing through Apple/Google and is far less likely to be flagged than a direct virtual-card checkout.
• Never initiate a chargeback against OpenAI or Anthropic — that is one of the fastest ways to guarantee a permanent ban.

The US$200 plans are higher-risk

ChatGPT Pro and Claude Max both cost US$200 a month, and both are more likely to be scrutinised: they sit on the risk team’s radar because a US$200 account with a suspicious payment method or usage pattern is a bigger fraud target. In practice that means (a) your payment anomalies are more likely to be escalated, (b) a ban loses you far more money, and (c) if the payment method was risky, the blast radius is worse. Start on ChatGPT Plus or Claude Pro (about US$20), run the account cleanly for a month or two, then upgrade.

What you put in the box matters (Claude especially)

Anthropic’s risk systems look at far more than the words you type. In July 2026, independent researchers reverse-engineered the Claude Code client and confirmed it embeds hidden markers for China-system timezones (Asia/Shanghai, Asia/Urumqi), China-related domains, Chinese AI-relay keywords, and custom upstream proxy URLs — the signatures of mainland relay and resale setups. The practical upshot for a HK or China user:

• Do not point Claude Code at a Chinese relay/API-proxy upstream (a custom ANTHROPIC_BASE_URL). That single setting marks the account as a relay user.
• Do not run your machine or browser on a China timezone while using the account, if you registered as a supported country.
• Avoid relay-style bulk workloads — large-scale translation batches, automated generation pipelines — which are exactly what resellers run.
• Do not ask it to build or explain tools for bypassing geo-restrictions — that lands straight in the abuse queue.

Phone number and account hygiene

• One person, one account. Multiple accounts on one phone number, one card, or one IP all look like bulk abuse.
• Do not reuse one verification number across many accounts — the number gets flagged and everything tied to it is at risk.
• Keep your verification number reachable for a few weeks so re-verification does not fail.
• Use a stable email (Gmail/Outlook/own domain). Temporary-mail addresses are a known KYC red flag at Anthropic.
• Turn on two-factor authentication — an account that is not yours any more is just a banned account waiting to happen.

Pacing and automation signals

Anthropic runs a 24–72 hour observation window after signup and after payment. Inside that window, keep usage light and natural: a handful of chats a day, no batch jobs, no API-key generation, no marathon sessions. Outside the window, avoid the patterns that trip automated-behaviour detection: sessions of many hours without a break, perfectly regular request intervals, very fast prompt bursts, and anything that scrapes the web app.

If you do get banned

• Do not immediately create a new account from the same IP, card or device — that looks like ban evasion and the new account is usually killed too.
• Appeal once through the official channels, honestly, and then move on. For Claude, appeals rarely succeed.
• Treat every ban as a learning event: the account that was killed was sending a signal you can fix before the next one.

The keep-it-alive checklist

• One consistent residential IP for signup, payment and every login.
• Same country across VPN, phone number, card and billing address.
• Own, unshared account; one verification number used once.
• Payment via App Store/Google Play or a clean, un-flagged card; never charge back.
• Start on the US$20 plan; upgrade to US$200 only after the account is stable.
• Light, natural usage for the first 72 hours; no relays, no scrapers, no marathon sessions.
• Stable email, 2FA on, and — for Claude — no China timezone or relay upstreams.