Credential isolation
A security design principle that fully separates sensitive credentials (such as payment credentials and passwords) from an agent's context and execution environment, ensuring that the agent cannot access the real credentials.
Core Facts
Timeline (last 90 days)
The key to security domain layering lies in not sharing credentials between domains, not in the number of domains.
Common means of reducing the blast radius include least privilege, credential isolation, and short lifetimes.
Claude Duet does not directly access users' passwords or login tokens; checks have been added to refuse sharing information such as credit card numbers and government ID documents; sensitive information such as health conditions and personal finances is only sent out when the user explicitly instructs it.
Hosted Deep Agents adds to the basic agent loop the capabilities required when a large number of users share the same agent: per-user isolated credentials and memory.
Credential isolation requires permission checks and audit logging before every tool call.
Permission governance frameworks for resident agents often include credential isolation (separating agent credentials from the user's primary credentials and using revocable dedicated tokens), sandboxed operation (limiting file system and network scope), and audit logs (recording every read and write operation).
All Facts (6)
Claude Duet does not directly access users' passwords or login tokens; checks have been added to refuse sharing information such as credit card numbers and government ID documents; sensitive information such as health conditions and personal finances is only sent out when the user explicitly instructs it.
65%UnverifiedThe key to security domain layering lies in not sharing credentials between domains, not in the number of domains.
50%UnverifiedCommon means of reducing the blast radius include least privilege, credential isolation, and short lifetimes.
50%UnverifiedHosted Deep Agents adds to the basic agent loop the capabilities required when a large number of users share the same agent: per-user isolated credentials and memory.
50%UnverifiedCredential isolation requires permission checks and audit logging before every tool call.
50%UnverifiedPermission governance frameworks for resident agents often include credential isolation (separating agent credentials from the user's primary credentials and using revocable dedicated tokens), sandboxed operation (limiting file system and network scope), and audit logs (recording every read and write operation).
50%