Anthropic's updated usage policy takes effect on November 12, ten days from now. When it was published in October, most of the attention went to the rule against abusing Claude. But the "do not abuse the platform" section also gained two other lines: a ban on unauthorized resale or proxying of Claude access, and a ban on using stolen, fraudulent or unauthorized payment methods to avoid paying.
We compared the page against Internet Archive snapshots. The June 1 capture still shows the old version, effective September 15, 2025, and that section has neither line. The October 14 snapshot already includes them. In other words, they entered the policy in the same October revision, and the discussion at the time simply focused elsewhere.
"Resell, proxy, or otherwise provide access to Claude through unauthorized means"
What the clause names
The resale clause is followed by two examples: services that relay requests through consumer subscriptions, and services that misrepresent the product or client they are using.
The first maps onto the familiar "account pool" model, in which an operator buys a batch of Pro, Max or Ultra subscriptions and repackages the web or Claude Code allowance as an API sold to others. Going by the wording, the second could cover setups that pose as the official client when sending requests. That is our reading; Anthropic has not spelled out the specific forms.
On the payment side, stolen cards and fraud are easy to understand, but "unauthorized" and "otherwise evade payment" reach much further. The policy gives no list, so whether gray channels such as third-party top-ups or gift cards count will only become clear through enforcement.
The old policy was not silent either. It already barred providing accounts or API access to people in violation of the supported-regions policy, and it barred helping banned users get around a ban. What changes now is that resale itself is a violation, even when neither buyer nor seller has any regional issue.
Chinese-speaking users are affected most directly
Mainland China, Hong Kong and Macao are not on Claude's supported list, and many individual developers and small teams get access through relay services, shared accounts and third-party top-ups. Over the course of this year, developer communities have repeatedly reported relay accounts being banned in batches and services dropping out without warning. Most of those accounts come from users and relay operators themselves, and the scale cannot be verified.
Once the new clause lands, a few things are worth thinking through in advance:
- Relay-service users: the clause binds the party doing the reselling or proxying, but your prompts and code still pass through a third party's servers first. If the provider runs out of working accounts, there is no official channel to back your balance or your deliveries.
- Those who sell accounts or run shared plans: until now they could argue the terms did not mention it. After November 12 that argument is gone.
- Teams building products: wiring Claude into your own features through the API or a cloud provider is a different thing from forwarding consumer subscriptions. Whether a reverse proxy for internal use counts as "providing access" is not stated in the text, and Anthropic has not publicly explained it.
Anthropic has also not said whether buyers' accounts will be penalized along with sellers'. The general terms allow warnings, rate limiting, suspension or termination of access, and how far it goes will depend on actual enforcement cases.
Two companies move in the same month
OpenAI also changed its rules last month. Its new version takes effect on November 20, and model routers and AI gateways that offer OpenAI models to third parties must pass a review first. The two dates are only eight days apart and the direction is similar, but the drafting differs: OpenAI sets up an approval step that leaves a door open for compliant gateways, while Anthropic's passage says only "unauthorized" and publishes neither a list of authorized parties nor an application path.
Anthropic's tug-of-war with third-party tools is not new either. In April it briefly cut off the path by which subscription allowances powered tools such as OpenClaw, then reopened it and listed agent allowances separately. That episode shows how the rules for "running third-party tools on a subscription" can swing back and forth within a year. Teams that depend on that path would do better to check their own integration now than to investigate after an account runs into trouble.
Anthropic has not disclosed how it checks for resale, nor how many bans of this kind it has issued. Both will have to wait for enforcement after the policy takes effect.
Sources: Anthropic Usage Policy (current text), Internet Archive snapshots, CocoLoop, reports on OpenAI's usage policy. The new clauses are based on a comparison of two archived versions; community reports of bans have not been confirmed by Anthropic.