Enterprise & Security

Claude Platform User Profiles: Enforcement Hits the End Customer, Not Your Whole Org

2 min read AI-generated

There are organization webhooks in the Admin API too. They fire on two events: when Anthropic restricts a user profile, and when that restriction ends.

Featured image for "Claude Platform User Profiles: Enforcement Hits the End Customer, Not Your Whole Org"

The Claude Platform release notes for October 20 carry three entries, and one of them fixes something platform operators have been complaining about for months.

User profiles: who is behind the API call

Until now, Anthropic only saw your organization on an API request. If a single end customer on your platform did something that triggered a safety enforcement action, the action landed on the API key—meaning on you, and on every other customer with you.

With user profiles you can now tell Anthropic which end customer a given request belongs to. An enforcement action then applies to that one profile instead of the whole organization. The new endpoints live under /v1/user_profiles, and both the SDKs and the ant CLI already support them.

This is aimed squarely at platforms, gateways and anyone reselling API access. If you ship Claude inside your own multi-tenant product, you were effectively liable for the aggregate behaviour of all your tenants.

Webhooks for bans and unbans

To go with it, the Admin API gets organization webhooks, in beta behind the webhooks-2026-10-07 header. Through /v1/organizations/webhooks you register endpoints for two event types: Anthropic restricts or bans one of your organization’s user profiles, and that restriction ends. Access runs through your Anthropic account team.

Together that gives you a flow without polling: create the profile, attach requests to it, and when Anthropic steps in, your system knows immediately which tenant is affected.

Inference hooks are out of beta

The third entry is short: inference hooks are no longer in beta for Claude Enterprise organizations. If you route prompts and responses through your own DLP or audit systems, you are now on a feature with regular support.

Multi-tenancy is a security feature, not a billing one

User profiles reads like a billing feature, and a year ago I would have filed it that way. It isn’t. For anyone reselling Claude, shared liability was the biggest uninsured risk in the product: one abusive customer could take the service down for everyone. Anthropic drawing that line itself, and reporting across it with webhooks, is the most concrete improvement for gateway operators since the Admin API shipped.

Sources

Claude PlatformAPIEnterprise