In brief: Agent Identity shifts access control from user-based to agent- and channel-based, enabling administrators to centrally define which tools, APIs, and data an agent may use in which context.
Anthropic introduces “Agent Identity” as a new access model for Claude Tag, equipping AI agents with their own workspace accounts rather than having them act as representatives of individual users. This addresses the growing autonomy of agents and the requirements of team-wide collaboration in shared channels.
The problem with classical user impersonation: In personal AI assistants, granting access works through linking personal accounts (Google Drive, GitHub, Calendar). For Claude Tag, where the agent works in shared Slack channels alongside multiple people simultaneously, this model breaks down: First, agent autonomy is growing — AI models double their ability to handle longer tasks autonomously approximately every four months. Agents schedule tasks for later, respond to events after the user logs off. Second, with multiple people in the channel, the question arises: whose access rights apply? There is no universally correct answer.
Agent Identity as a solution: The agent receives its own workspace accounts in every system it touches — posts as “Claude App” in Slack, creates pull requests as “Claude GitHub App”, queries data warehouses via service accounts. This prevents personal user credentials from circulating in shared channels and thus serving as a backdoor to private documents.
Implementation at admin level: Administrators define a baseline identity at workspace level with accounts, tools, API keys, and permissions. Each channel inherits this identity by default, but can override it — for example, giving an engineering channel warehouse access while a CRM connector is available only in a private channel. This includes: repository access (which repos are readable/writable), connectors (tools and API keys with permission levels such as read-only vs. write), skills and plugins (dynamically loadable instructions for specialized tasks), and standing instructions (custom prompts per channel).
Administrative advantage: Because Agent Identity operates around unique agent identities, revoking an identity ends access everywhere it was used. This requires significantly less audit effort than tracking individual agent actions across dozens of user accounts. The access logic thus shifts from “What may this user do?” to “What may this agent do in this channel?”
Source: claude.com · Published 23 June 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.2.