Towards Superintelligence Alignment
lensframework

Control protocol

Monitors and safety rates around an untrusted model under intentional subversion — not a moral target.

Lenses and frameworks

Role. Protocol. Graph node control.

AI Control studies protocols around untrusted models under intentional-subversion threat models. Its monitor games and safety rates are protocol evidence. They do not supply a moral target.

— Frames draft §4.9 (Greenblatt et al. 2024)

A protocol family for deployment when the objective is untrusted. It composes as a leaf of a safety case. It does not replace the correction relation.