Anthropic launches Claude for Enterprise

Background

Organisations that use a large body of internal knowledge, such as company documents, code and records, need an assistant that fits their security and administration rules. A chat product designed for individuals does not give an administrator the controls that a company needs to manage who uses it and what happens to its data. Claude for Enterprise was Anthropic's plan for that group of customers.

What happened

In September 2024 Anthropic announced Claude for Enterprise, a plan for organisations working with internal knowledge.^[1] The announcement set out the context window, the security features and the early customers that the company named.

Key details

The plan offered a 500K-token context window. A context window is the amount of text, measured in tokens, that the model can take into account in one conversation. A larger window lets a user give the model a long document or a large body of material at once.^[1]

Security features included single sign-on (SSO) with domain capture, role-based permissions, audit logs and SCIM provisioning. Single sign-on lets people log in with the company's own identity system. Domain capture ties accounts that use a company's email domain to the organisation. SCIM is a standard way for an identity system to create and remove user accounts automatically. Role-based permissions give different people different levels of access, and audit logs record activity for later review.^[1]

A native GitHub integration, in beta for early Enterprise users, synced code repositories with Claude.^[1] The integration let the model work with code that already lived in an organisation's repositories, rather than with copies pasted into a chat.

GitLab and Midjourney were named as early customers.^[1] Anthropic also stated that it does not train Claude on customer conversations and content.^[1]

Impact

Claude for Enterprise was a step in Claude's move from a general chat product toward a tool that an organisation can govern. The combination of a large context window, identity and access controls, audit records and a code integration addressed the concerns that come with putting internal knowledge into an AI system.^[1]