How Not to Lose Your Working Claude: Breaking Down Anthropic's Ban Waves

Anthropic has spent years building its reputation as the most ethical AI company, yet in the second half of 2025 alone it banned 1.45 million accounts. In 2026, researchers recorded several more waves of mass bans.

In this article, we break down the ban cases and share tips to help you avoid losing access to your data after a ban. 

What We Know About the Claude Ban Waves

On January 29, 2026, Anthropic published the Transparency Hub, which gives the community a wealth of interesting figures. The report covers 1.45 million banned Claude.ai, API, and Code accounts. Of the 52,000 appeals submitted, only 1,700 were successful. 

Anthropic noted that the vast majority of bans were triggered by auto-detection of spam, abuse, and ToS violations. Judging by the number of successful appeals, the chances of recovering access are minimal — even when the ban is a mistake. 

As of April 2026, four waves of bans had been recorded on Claude. Each was driven by a key trigger: from sanctions policy to cracking down on limit abuse.

In September 2025, Anthropic banned the use of its products for companies with majority ownership from China, Russia, Iran, and North Korea. The company acknowledged this would cost hundreds of millions of dollars in revenue. The result was mass bans of VPN users from mainland China and other sanctioned regions — even when accounts were paid for with a USD card. 

The second wave was a mass outage in March. The Claude website, Console, and Code all went down simultaneously. Thousands of users received the message "Your account has been disabled after an automatic review," even though these were technical failures, not actual bans. 

Notably, after the service was restored, some users continued to receive account suspension emails — the auto-detection had flagged unusual activity patterns. Anthropic later acknowledged the error, but appeal processing times ran at 2–3 weeks. 

The third wave was the strangest. In April 2026, Anthropic began flagging certain users as minors. Over 400 researchers signed an open letter against the policy — the main concerns being biometric data collection and the risk of data leaks. Users on Reddit and X reported that accounts with paid subscriptions were being banned as well.

Another large-scale case involved the banning of hundreds of work accounts at an IT company. The story was covered on Hacker News. 110 accounts were blocked at once, without warning or explanation — while charges continued to be billed through the API. 

How to Reduce the Risk of a Ban?

The most important thing to understand is that the official appeals process barely works. If you end up in the 96.7% rejection pool — you won't get your money back, and you won't get your account back either. The strategy, therefore, is built around not tying your entire infrastructure to a single account. 

Split your billing between API and subscription. Pro and Max are one system; API via Console is another. If your subscription gets banned, the API keeps working. This is the cheapest insurance available: a separate API key via console.anthropic.com, billed per token. 

Don't pay for your subscription using gift codes or VPN. This is the most common auto-detection trigger. A USD card from a trusted GEO plus a static IP equals the minimum number of signals for a block. If you have to use a non-standard payment method — create a separate account for it.

Don't run OpenClaw or other agents on top of Pro/Max. As of April 4, 2026, this is a formal grounds for a ban. Either pay through the API or keep such tools on a separate account you can afford to lose.

Save your projects and configs locally. Claude Code sessions are stored in ~/Library/Application Support/Claude — from there they can be manually transferred to a new account after a ban. Everything stored in Anthropic's cloud — Claude Design, Cowork sessions, and so on — disappears along with the account and cannot be recovered. 

Maintain parallel access to another provider. GPT-5, Gemini, Mistral — any of them will cover basic tasks while you wait on an appeal. This is a basic operational fallback for teams whose workflow depends on LLMs.

Conclusion

It appears that bans are an architectural decision on Anthropic's part. It's cheaper for the company to restrict accounts in bulk via auto-detection than to review cases manually. This approach may change in the future.

The advice for webmasters is clear: don't rely on a single provider — a parallel API via GPT-5 or Gemini eliminates most of the risk. A separate account for experimenting with agents and third-party tools, local backups of projects and configs — this is the bare minimum that lets you survive a ban.