Claude API resellers can now get one customer banned instead of their whole account
Companies that resell Claude or build apps using it can now tell Anthropic which customer each API request belongs to. If a customer breaks the rules, Anthropic can ban that customer without banning the company's entire account, Anthropic says.
Anthropic's image for the Claude Platform, where the new user profiles are available. Image: Anthropic
Before this change, one customer's actions could put an entire company at risk. Anthropic uses automated safety checks to review API traffic. Its documentation says that if a customer does not have a profile, their actions can lead to enforcement against the whole organisation. If Anthropic bans a customer who has a separate profile, the company's other API requests are not affected.
Each customer gets a profile ID that goes with every request
Developers register each customer once and save the ID they receive. They then include that ID in a header with every request made for that customer. Profiles are free and do not affect prices.
There are two types of profiles. An app creates one profile for each person who signs up. A gateway that resells API access creates one profile for each company it serves, rather than one for every person at that company. Requests without a profile count as the organisation's own usage. Anthropic recommends keeping that usage in a separate organisation.
A profile contains only the customer information that the developer provides. It does not contain conversation content. Anthropic does not check whether the information is correct. Profiles work with the Claude API, Claude Platform on AWS, and Microsoft Foundry deployments hosted by Anthropic. Amazon Bedrock and Google Cloud do not support the profile endpoints. Developers using Claude through those services must create profiles through the Claude API instead.
Artwork from Anthropic's enterprise page. Profiles are designed for companies that sell Claude to their own customers. Image: Anthropic
New webhooks tell organisations when a customer is banned
Anthropic has also introduced organisation webhooks in beta. These send two kinds of notifications: one when Anthropic limits or bans a customer profile belonging to an organisation, and another when that restriction ends. Organisations can request access through their Anthropic account team.
Profiles are optional. Platforms that do not use them still face the old risk: one customer's bad behaviour could affect everyone. This change follows the release of other administrative tools, such as a feature that lets organisations retrieve the instructions their members entered into Claude.