Hot AI News

Claude’s linked billing lets app users pay Anthropic directly, with no API key shared

Anthropic's new beta lets customers of third-party apps pay for Claude usage through their own Claude Platform organization, with a spend cap, instead of handing over an API key.

Claude’s linked billing lets app users pay Anthropic directly, with no API key shared
Image: Anthropic

Anthropic has put linked billing in beta on the Claude Platform. Users of tools such as Factory, Linear and Ramp can now pay for Claude API usage through their own Claude Platform organization, without sharing an API key.

How the customer-pays flow works

According to Anthropic’s release notes and developer documentation, a customer links their organization to an app once. That creates a connection. After that, the app chooses on each request who pays, by setting billing_policy on calls to /v1/messages or /v1/chat/completions. The field takes bill_to: "user_profile" for the customer or bill_to: "self" for the app. The beta header is user-profiles-billing-2026-10-01.

Claude's linked billing lets app users pay Anthropic directly, with no API key shared
Image: Anthropic

The person approving the link must hold the Admin or Billing role in the customer’s organization. They set a monthly spend limit during approval. In the Claude Console, customers can also set a spend cap for the app or unlink it at any time. The docs say that if a request marked for the customer cannot be billed to them, the API refuses it rather than billing the developer.

Advertisements

Who is eligible and what still does not carry over

The feature works only for customers whose organization pays Anthropic directly. Prepaid credits and invoiced billing both count. Organizations billed through AWS, Google Cloud or Azure marketplaces, and those on only a Claude subscription plan, are not eligible and get a 403 user_profile_not_billable error. Monthly API credits bundled with Max, Ultra and Team plans, and credits from partner or startup programs, cannot pay for these requests.

Requests still run in the developer’s organization. The developer’s rate limits, model access, data retention and encryption keys apply, not the customer’s. The documentation says zero data retention, BAA and CMEK arrangements held by the customer do not apply, and it advises against linking customers who need HIPAA, FERPA or CMEK coverage for this usage. Message Batches cannot bill the customer, and billable code execution features are rejected on customer-billed requests.

Setup is not trivial for developers. Anthropic issues no client secrets, so an app’s back end authenticates through a registered identity provider and Workload Identity Federation. Until an app is published, only designated test users can link it. Connection access tokens last up to an hour, and refresh tokens at most a year.

Fewer shared keys, but trust moves to the app

The analysis angle here is rules and safety. The obvious gain is that a customer no longer pastes a long-lived key into a third-party tool. That matters: Railway put a $740K price on the October hack that hunted AI keys, according to our earlier report. A stolen key is a standing credential, while a linked connection can be revoked in the Console and is capped by a spend limit the customer sets.

The design also leaves some risk in place. The docs say the Claude API cannot verify that a developer bills a customer only for work the customer asked for. Anthropic states that testing, evaluations or work for other customers must not be charged to a linked customer, and that this is the developer’s responsibility. The docs add that a request naming a different linked profile bills that profile’s organization with no error. The connection also belongs to the whole organization, not the individual who approved it, so it stays active if that person leaves. Anthropic says a completed link does not guarantee billing will work, and tells developers to say “linked” rather than “billing active”.

Rivals approach the same problem differently. Perplexity’s Orchestrator API puts 85+ models behind one key, which simplifies access for a developer but still centers on a key the developer holds. Anthropic’s route instead moves the bill to the end customer’s own account.

Advertisements

What the sources leave unsaid

The announcement is a beta. The docs warn that endpoints, parameters and behavior can change and that it may not suit production use. The sources do not say which of Factory, Linear or Ramp have enabled it, nor give pricing beyond standard API rates, which the customer’s organization pays. Anthropic’s claims about how the feature behaves have not been independently verified. Usage reporting in the Console and the Usage and Cost Analytics API is per profile but not real time.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like