Claude API resellers can now get one customer banned instead of their whole account
Companies that resell Claude or build apps on it can now tell Anthropic which end customer each request belongs to. If one customer breaks the rules, a safety ban can then hit that customer instead of the whole account, Anthropic says.
Anthropic's share image for the Claude Platform, where the new user profiles live. Image: Anthropic
Until now that was a real risk. Anthropic runs automated safety classifiers over API traffic, and its documentation says that without a profile, one end customer can trigger an enforcement action against the whole organisation. When a ban lands on a single profile, the rest of the traffic is unaffected.
Each end customer gets a profile ID that travels with every request
Developers register each customer once and store the ID that comes back. They then send that ID in a header on every request made for that customer. Profiles are free and do not change pricing.
There are two kinds. An app creates one profile for each signed-up user. A gateway that resells API access creates one for each company it serves, not for each person at that company. Requests without a profile count as the organisation's own usage, and Anthropic advises keeping that in a separate organisation.
A profile holds only what the developer sends about the customer, not conversation content. Anthropic does not verify those details. Profiles work on the Claude API, Claude Platform on AWS and Microsoft Foundry deployments hosted by Anthropic. Amazon Bedrock and Google Cloud do not offer the profile endpoints, so developers who run Claude there create profiles on the Claude API instead.
Artwork from Anthropic's enterprise page. Profiles are meant for companies that sell Claude to customers of their own. Image: Anthropic
New webhooks tell an organisation when one of its customers is banned
Anthropic also launched organisation webhooks in beta. They send two events: when Anthropic restricts or bans one of an organisation's user profiles, and when that restriction ends. Access is by request through an Anthropic account team.
Profiles are optional. A platform that skips them keeps the old risk that one bad customer takes everyone down. The change follows other admin tools, such as letting organisations pull the instructions members typed into Claude.
