Claude Account Bans, Suspensions and Appeals

Reuters reports that Anthropic's new anti-distillation defense cut stolen reasoning tokens by 90%–99%, banned tens of thousands of accounts, and named Zhipu, saying GLM-5.3 took cybersecurity capabilities from Claude during training.

A user lists about 300 yuan a month in AI subscriptions, roughly 3% of base salary, with the $20 Claude plan stopped after the account was banned.

One observer says the Claude ban wave has spawned a full gray-market toolchain of one-click cleanup, fingerprint rotation and environment checks, and argues the root cause is Anthropic's account system not fitting mainland China.

An analysis notes that Anthropic's sub-0.1% false-positive figure refers to its request classifier rather than wrongful bans of ordinary accounts, and that the article gives no account-level false-ban data.

A user publicly asks to borrow someone else's Claude account because their own account has been banned.

A breakdown of Anthropic's anti-distillation post: new measures cut attacker data collection by roughly 90-99%, tens of thousands of related accounts were banned, and users should check what their prompts and tools ask Claude to reveal.

A user says their Claude account was banned without a refund and names an Anthropic staff member in the complaint.

A user jokes about how to price ban insurance: by ban probability or by balance, with a "revive" plan that pays out a new account per ban.

An explainer says subscribing to Claude Pro does not protect an account, since Anthropic can ban users for violations, unsupported-region signups or terms breaches, and circulating anti-ban tips are not official rules.

A post uses Claude's account suspension notice to teach the word "suspend," with an image joking about the wave of bans.

A user who says their account has run for years without a ban credits an unchanged timezone and language, a self-hosted VPS with a dedicated IP, and usage that rarely maxes out.

The poster says electricity, internet and family costs outweigh the benefit, and that risk controls track behavioral fingerprints rather than location, so a clean device and normal use are the real answer.