PureIP

Anthropic's Claude ban numbers, three reports side by side: 11.4M accounts banned in H1 2026, appeal overturns up to ~10.6%

2026-11-16 · figures from Anthropic's Transparency Hub; ratios computed by us and labelled · by PureIP · 中文

Claude ban numbers get quoted all over X, usually for one period and without a source. Anthropic publishes banned accounts, appeals and appeal overturns twice a year in its Transparency Hub. Below are three reports side by side, each figure traceable to the official page or an official PDF. A long X post we checked quotes 11.4 million, 398k and 42k for H1 2026 and 1.45 million for H2 2025; those match the official figures.

The three official reports

Anthropic Transparency Hub: banned accounts, appeals, appeal overturns (as displayed on the official page)
PeriodBanned accountsAppealsAppeal overturns
Jan–Jun 2025690k35k1.0k
Jul–Dec 20251.45 million52k1.7k
Jan–Jun 202611.4 million398k42k
690k 1.45M 11.4M H1 2025 H2 2025 H1 2026 Banned accounts, linear scale, official figures

The H1 2026 figures come from the page System Trust and Reporting, which says Last updated July 23, 2026. The earlier two periods come from official PDFs linked from the same page.

Ratios we computed

These are our calculations from the rounded official figures, not official statistics. Expect rounding error; read them for order of magnitude and direction only.

Computed by PureIP, for reference only
PeriodBans vs previous periodAppeals ÷ bansOverturns ÷ appealsOverturns ÷ bans
Jan–Jun 2025—~5.1%~2.9%~0.14%
Jul–Dec 2025~2.1×~3.6%~3.3%~0.12%
Jan–Jun 2026~7.9×~3.5%~10.6%~0.37%

What the numbers do not say

  1. No breakdown by reason. The report text says violations of the Usage Policy, the Consumer or Commercial Terms (as applicable) or the Supported Region Policy may lead to a warning, suspension or termination. The H1 2025 text mentions only the Usage Policy; the later two add the terms and the Supported Region Policy. That is a change in wording, not evidence that region-related bans changed in share.
  2. No breakdown by product or country. claude.ai, the API and Claude Code are not separated, so you cannot derive "how many bans in region X".
  3. No definition. The page text we read does not define "banned accounts". How many are bulk-registered spam or automated abuse is not stated, and we make no inference.
  4. No outcomes other than overturns. Rejected, pending and withdrawn appeals are not counted, so 100% minus 10.6% is not a rejection rate.
  5. No predictive value for one account. These are half-year totals across all accounts.

How to verify

This page only sets out official data. It does not offer or recommend ways around regional restrictions or risk controls, and a network check does not change where you are or whether you are eligible to use a service. Check what your current egress looks like to AI services.

PureIP is a free online toolbox for network checks: IP reputation, AI-service reachability, global ping, DNS and WHOIS. Runs on Cloudflare Workers, no signup.

Results reflect network-layer and public data only. They say nothing about any account's status and do not represent any decision by Anthropic. Corrections: contact@mail.pureip.app.

IP looks risky? VPS providers we actually use: RackNerd · 搬瓦工 · DMIT · Full comparison → Affiliate links: we may earn a commission if you buy through them, at no extra cost to you.