Connector permissions are the access rights granted to a service linked to another system, such as an AI assistant connected to a code repository. They define what the connector can do-for example, read files, modify content, or share information with other services.
These permissions matter because a compromised or misused connector can expose data or make unauthorized changes. In attacks involving indirect prompt injection, untrusted text in connected content may influence an assistant with tool access. Limiting a connector to the minimum access needed, separating read and write capabilities, and requiring human approval for consequential actions can reduce risk. Regularly review connector scopes and audit logs, especially when repositories contain secrets or sensitive code.