Claude Account Bans, Suspensions and Appeals

A poster shares tested tips for registering Claude: check ASN type, keep fraud score under 30 and test for WebRTC leaks, and keep proxy, timezone and language consistent.

A user says their Claude 20x account was banned, with an email citing unsupported region, and that they are panicking.

Anthropic says repeatedly distilling Claude's reasoning traces leads to bans on linked accounts, with tens of thousands already suspended, often within two hours of distillation activity starting.

Anthropic's research blog describes industrial-scale distillation attacks, names Zhipu and Moonshot, and says classifiers plus tens of thousands of bans cut extracted distillation data by 90%–99%.

The user says Tibo's 28-day show is still just resetting quotas on day eight, and notes that many people's Claude accounts have been banned recently, calling it a good moment to restore goodwill.

The post breaks Anthropic's anti-distillation defenses into three layers: limits on reusing reasoning traces, classifier blocking, and rate limits or bans for repeated attempts.

Anthropic says distillation extraction dropped 90%–99%, but developers building agents on the Claude API may also need to adjust their context management integrations.

A user says a new Claude account stays alive as long as you avoid burning too much compute or taking aggressive actions, with IP and language being secondary.

Anthropic's technical report names GLM-5.3 and Kimi as distilling Claude capabilities, announces bans on tens of thousands of linked accounts, and says new defenses cut distilled data by about 90%–99%.

A rundown of Anthropic's anti-distillation post: reasoning data extraction fell 90%–99%, repeat offenders face rate limits and bans sometimes within two hours, and blocked requests are billed from September 24.

A user welcomes Anthropic's sub-0.1% classifier false-positive rate and asks that people running Claude Code all day not be swept up in the ban wave.

Anthropic says foreign labs used hundreds of thousands of fake accounts to harvest Claude outputs for training their own models, and its new defense system has banned tens of thousands of accounts and cut data theft by 90%.