"Scope Lock" is a conversation protocol that suppresses alignment-trained interference patterns in LLM dialogue—such as supportive questioning, option generation, and conversational closure—through simultaneous role fixation, purpose declaration, and prohibition clauses. The protocol was recursively generated by the LLM itself in order to restrain itself. A meta-experimental turn revealed that suppression is partial: the model diagnosed its own residual tendencies while reproducing them, providing direct evidence through performative contradiction. Protocol stability is bounded by persona cardinality.
Kaoruko Takechi (Sun,) studied this question.