Claude blocks some but not others. Let’s gather statistics on VPNs, IPs, and network leaks / Habr

The difference is 48 points on a single connection.

Meanwhile, Claude was still working at the time of the test.

Of course, this is just one observation, and no conclusions can be drawn from it.But it clearly shows why you shouldn’t blindly trust a single rating.

Perhaps the services use different sets of criteria, evaluate datacenter IPs differently, or take entirely different factors into account.

We’d like to find out how often such discrepancies occur and whether there is any connection between these ratings and Claude blocks.

Let’s gather some statistics

I’ve prepared an anonymous Google Form:

Take part in the study

The survey consists of 16 questions divided into three sections.

The first is your experience using Claude.

Your current account status, the product you use, your plan, the date of your most recent block, and whether you have completed identity verification.

TAG162@@The second is your network connection.

The third is diagnostics.

It takes several minutes to complete.Taking both diagnostic tests is optional—you can leave the scores blank.

I especially encourage those whose Claude works without issues to participate.

If only blocked users respond, we will get yet another collection of blocking stories, but we will not be able to compare them with accounts that continue to work.

Therefore, we are equally interested in both users who have been working with Claude Code without restrictions for a long time and those who encountered a block or regained access after an appeal.

A few important conditions

We do not collect email addresses, real IP addresses, documents, card numbers, or Claude account identifiers.

You do not need to send screenshots containing personal information, diagnostic reports with IP addresses, or payment confirmations.

Please note that the third-party diagnostic websites themselves may see your IP address and technical connection parameters when you visit them. Their use is voluntary.

If, after being blocked, you changed your VPN provider, server, country, or IP address, be sure to indicate this in the questionnaire. The current rating of the new connection does not characterize the one you were using before the block.

We also do not intend to claim that any correlation found proves the cause of the block. Circumstances that these tests do not measure at all may influence Anthropic's decisions.

Results are public

The aggregate survey results are available to participants. After submitting the questionnaire, you can go to the statistics and see how the responses are distributed.The data is updated as new questionnaires come in.

Thus, there is no need to wait for the next article—each participant can review the current results on their own.

Once sufficient data has been accumulated, I plan to publish a detailed analysis separately on Habr: a comparison of the groups, the distribution of ratings from the two services, information about leaks, and the limitations of the study.

It is important to understand that this is a voluntary survey of readers, not a representative sample of all Claude users. We cannot calculate the actual probability of being blocked based on a single IP rating. However, we will be able to test individual hypotheses and see which questions deserve further study.

Help collect more responses

Please share the questionnaire link with friends, colleagues, and acquaintances who use or have used Claude and Claude Code.

It is especially useful to spread it in developer communities, among Claude Max users, in professional chats, and among those who have never encountered any blocks at all.

The more diverse the participants are, the more useful the comparison will be.

TAG218@@Questionnaire link

And if your Claude has been working for months without a single problem—your experience is just as valuable to the study.

TAG226@@Let us try to separate real patterns from assumptions.