The Collaboration Protocol: Silent Intent Inference as an Accountability Challenge in Human-LLM Interaction

Human-AI collaboration challenges persist even when AI systems are technically capable. We argue these challenges may often stem from Silent Intent Inferencea structural pattern where AI resolves semantic ambiguity without explicit user authorization. We propose framing this not as mere interaction friction but as a potential accountability challenge: it may create oversight gaps, dicult-to-trace decisions, and gradual reduction of user agency. We introduce the Collaboration Protocol (CP), a theoretical framework for preserving human semantic authority through four operationalized constructs: (1) Intent Control Degree (ICD), a user-adjustable parameter (0.01.0) governing how ambiguity is resolved; (2) Intent Pivot Points (IPP), semantic locations where meaning may branch, with a six-type taxonomy; (3) Intent Fidelity Index (IFI), a composite metric measuring intent preservation across interactions; and (4) Context Roots (CR), a mechanism for anchoring ambiguity in interaction history, enabling multi-turn and multi-session intent tracking. The framework introduces the Semantic Materiality Proxy (SMP) for objective ambiguity classication and the Context Factor (CF) for modeling how prior interactions aect current ambiguity. Each construct includes specic measurement procedures enabling empirical validation. We propose a phased research agenda: quantifying silent inference prevalence in current LLM systems, validating ICD eects on collaboration quality, and establishing IPP annotation reliability. This paper establishes a falsiable theoretical foundationthe constructs and hypotheses are designed to be empirically tested and potentially refuted by the research community.