The approach focuses access decisions on the requested resource, the caller and relevant conditions. Confirming identity and deciding which actions are permitted are separate parts of the decision.
For an agent, a familiar account or private network should not automatically grant broad access. Trusted controls need to enforce the authorized operations; the model's willingness to follow instructions is not the security boundary.
ELI5
Think of checking both a visitor's identity and which rooms they may enter, rather than giving access to every room after they reach the building.
For example, an agent permitted to read a project document would still need separate authority to change a sensitive account setting.
Is a private network enough to establish trust?
No. Network location alone should not establish permission to access a protected resource.
Is zero trust a promise that incidents cannot occur?
No. It is an approach to access decisions that still requires sound implementation and operation.

