Understanding Claude Tag’s access model in Slack and how to configure it securely

Tenable Research explains Anthropic’s Claude Tag for Slack — a multiplayer AI agent that acts under admin-configured shared access bundles rather than individual user credentials — and analyzes the security implications. The report details the agent’s identity layers (installation, access bundles, and scopes), warns that channel membership becomes an access-control surface, documents coordinated disclosure with Anthropic, and provides four practical recommendations for administrators: audit bundles, review channel membership and bundle scope together, scope bundles minimally and manage invite policies, and prefer per-user delegation where appropriate.