BREAKING
Claude Tag adds 'agent identity'
DM vs Channel access
DM (one-on-one)
User
●
Uses the user's own credentials
●
Inherits user permissions
●
Logs are per user
Channel (multiplayer)
Agent
●
Claude's own identity
●
Defined at workspace, per-channel override
●
Logs are per Claude identity
Why 'act as user' broke down
What gets managed
1
Repo access
↓
2
Connectors
↓
3
Skills/plugins
↓
4
Instructions
Security controls
1
Allowed hosts only
↓
2
Per-identity audit logs
↓
3
Native system logs
↓
4
Single revoke
A new model for AI security
AI NEWS BLITZ
Anthropic gives Claude its own credentials in shared channels.