Osegi
AI·2 min read

Claude account blocked? Why the VPN theory may be wrong

Claude stops working and search results blame your VPN, your Russian IP or your card. A payment team that handled 15,000 subscriptions says the real picture is messier.

A person at a laptop looking at an error message while a VPN app shows connections to several different countries.
Image: Хабр

Your Claude stops working. You search, and every page offers a confident culprit: the VPN, a Russian IP, the Russian language, the card. Then a friend with almost the same setup carries on as normal, and the neat explanation falls apart.

The team at Raketa Pay, a service that pays for foreign subscriptions on behalf of customers, has looked at this a lot. They say more than 15,000 subscriptions have gone through them. That is not 15,000 investigated bans, and they have no view into the platforms' internal anti-fraud systems. Still, their argument is worth a read: no single parameter explains a block, the whole history does. That means the connection, the account, the payment and how the service was used afterwards.

Is the Claude ban rule actually known?

Not really. Media stories about Anthropic blocks describe a community theory about a points system, and Anthropic does not disclose how it spots violators. One expert quoted in the press cites company documents that supposedly mention Russian-language code comments as a criterion, but no document was shown. Seeing a pattern is not the same as knowing the algorithm.

The team also separates problems that users lump together under "ban". A card that fails, a network error and a disabled account are different things. OpenAI, for instance, documents a network error that carries a block message, and that alone does not mean the account is gone. Change the VPN, device and card all at once and you can no longer tell which change mattered.

Does a VPN cause blocks?

Not necessarily, but switching constantly may not help. OpenAI says unusual login locations, devices and activity can trigger warnings, though a warning is not a violation. Its advice for diagnosing suspicious activity includes using a trusted network and turning off VPNs or proxies. A static IP keeps one parameter steady, but it does not satisfy the service's other requirements, and a pricey VPN is no guarantee either.

Language, physical location and exit IP country are also separate things. The team found nothing in public rules that bans you for writing in Russian. Regional limits, however, are stated outright: OpenAI warns of suspension for access from unsupported countries, and Anthropic has its own supported-regions policy that also considers organisations' circumstances.

The team feels Claude cases are harder to untangle, but says that is an impression from support requests, not a controlled comparison. And what worked on ChatGPT proves nothing about Claude.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:claudechatgptanthropicopenaivpnaccount ban

More from this day

See all →

More in: AI

See all →